Ink Markup Language

regularChannels? intermittentChannels?

The <traceFormat> element describes the format used to encode points within <trace> elements. In particular, it defines the sequence of channel values that occurs within <trace> elements. The order of declaration of channels in the <traceFormat> element determines the order of appearance of their values within <trace> elements.

Regular channels appear first in the <trace>, followed by any intermittent channels. Correspondingly, the <traceFormat> element contains a <regularChannels> section followed by an <intermittentChannels> section. If no channels of a specific type exist, the corresponding element may be omitted.

3.1.2 <regularChannels> element

Attributes:

Contents:

The <regularChannels> element lists those channels whose value must be recorded for each sample point. The order of the channel declarations within the <regularChannels> element specifies the order in which the channel data samples appear within <trace> elements which use this format.

3.1.3 <intermittentChannels> element

Attributes:

Contents:

The <intermittentChannels> lists those channels whose value may optionally be recorded for each sample point. As with the <regularChannels> element, the order of the enclosed channel declaractions gives the order of the intermittent channel data samples within traces having this format.

3.1.4 <channel> element

Attributes:

Contents:

Within a <regularChannels> or <intermittentChannels> element, channels are described using the <channel> element, with name, type, and default attributes.

The required name attribute specifies the interpretation of the channel in the trace data. The following channel names, with their specified meanings, are reserved:

The type attribute defines the encoding type for the channel (either boolean, decimal, or integer). If type is not specified, it defaults to decimal.

A default value can be specified for the channel using the default attribute; the use of default values within a trace is described in the next section. If no default is specified, it is assumed to be zero for integer and decimal-valued channels, and false for boolean channels.

Typically, a channel in the <traceFormat> will map directly to a corresponding channel provided by the digitizing device, and its values as recorded in the trace data will be the original channel values recorded by the device. However, for some applications, it may be useful to store normalized channel values instead, or even to remap the channels provided by the digitizing device to different channels in the trace data. This correspondence between the trace data and the device channels is recorded using a <mapping> element within the <channel> element.

The <mapping> element can specify the identity mapping, or a formula expressed in MathML, or a lookup table. For a detailed description of the types of mappings supported by the <mapping> element and its usage, see the Mappings section.

If no mapping is specified for a channel, it is assumed to be unknown.

3.1.5 Orientation Channels

The channels Tx, Ty, A, E and R are defined for recording of pen orientation data. Implementers may choose to use either pen azimuth A and pen elevation E, or alternatively tilt angles Tx and Ty. The latter are the angles of projections of the pen axis onto the XZ and YZ planes, measured from the vertical. It is often useful to record the sine of this angle, rather than the angle itself, as this is usually more useful in calculations involving angles. The <mapping> element described in the Mappings section can be employed to specify an applied sine transformation.

The third degree of freedom in orientation is generally defined as the rotation of the pen about its axis. This is potentially useful (in combination with tilt) in application such as illustration or calligraphy, and signature verification.

Figure 2a displays the pen orientation using Azimuth and Elevation. The origin of the Azimuth is at the Y-axis. Azimuth increases anticlockwise up to 360 degrees. The origin of Elevation is located within the XY-plane. Elevation increases up to 90 degrees, at which point the pen is perpendicular to the XY-plane.

Figure 2b explains the definition of the Tilt-X and the Tilt-Y angles. For both the origin is along the Z-axis. Tilt-X increases up to +90 degrees for inclinations along the positive X-axis and decreases up to -90 degrees for inclinations along the negative X-axis. Respectively, Tilt-Y is defined for pen inclinations along the Y-axis.

Figure 3a displays the pen orientation decomposition as functions of Azimuth/Elevation or alternatively as function of Tilt-X/Tilt-Y. Thereby, Elevations of the pen which are mapped to the XZ- and to the YZ- plane lead to Tilt-X and Tilt-Y.

Figure 3b shows the Rotation of the pen along its longitudinal axis.

3.1.6 Time Channel

The time channel allows for detailed recording of the timing information for each sample point within a trace. This can be useful if the digitizing device has a non-uniform sampling rate, for example, or in cases where duplicate point data is removed for the sake of compactness.

The time channel can be specified as either a regular or intermittent channel. When specified as a regular channel, the single quote prefix can be used to record incremental time between successive points. Otherwise, the value of the time channel for a given sample point is defined to be the timestamp of that point in the units and frame of reference specified by its corresponding <captureDevice> description (more precisely, by the <channelDef> element for the channel).

As with the other predefined channels, the meaning of the integer or decimal values recorded by the time channel in a given trace is defined by the <captureDevice> information associated with the trace's traceFormat. In the case of the time channel, its <channelDef> element contains both a units and relativeTo attribute.

The units attribute gives the units of the recorded time values, and the relativeTo attribute describes the frame of reference for those recorded values. The value of the relativeTo attribute can either be an xsd:dateTime or xsd:time which gives the base timestamp for the time channel values in every trace, or it can have the value "trace", which means that the time channel values are relative to the beginning timestamps of the individual traces in which they appear.

The following example defines a time channel whose values for a given point are the timestamp of that point in milliseconds since midnight, January 1, 2003, UTC:

<channelDef name="T">
   <representation type="integer" units="ms" 
                   relativeTo="2003-01-01T00:00:00Z"/>
</channelDef>

This <channelDef> element defines a time channel whose values are the timestamp in milliseconds for a particular point offset from the beginning timestamp of the trace (see the section timestamps section for a description of trace timestamping):

<channelDef name="T">
   <representation type="integer" units="ms" relativeTo="trace"/>
</channelDef>

If no <captureDevice> information is provided, or if no value is specified for the relativeTo attribute, the ink processor cannot make any assumption about the relative timing of points within different traces. Likewise, if no units are specified, no assumption can be made about the units of the time channel data.

3.1.7 User Defined Channels

In addition, user-defined channels are allowed, although their interpretation is not required by conforming ink markup processors.

3.1.8 Specifying Trace Formats

The following example defines a <traceFormat> which reports decimal-valued X and Y coordinates for each point, and intermittent boolean values for the states of two buttons B1 and B2, which have default values of "false":

<traceFormat id="xyb1b2">
   <regularChannels>
      <channel name="X" type="decimal">
         <mapping type="identity"/>
      </channel>
      <channel name="Y" type="decimal">
         <mapping type="identity"/>
      </channel>
   </regularChannels>
   <intermittentChannels>
      <channel name="B1" type="boolean" default="F">
         <mapping type="identity"/>
      </channel>
      <channel name="B2" type="boolean" default="F">
         <mapping type="identity"/>
      </channel>
   </intermittentChannels>
</traceFormat>

The appearance of a <traceFormat> element in an ink markup file both defines the format and installs it as the current format for subsequent traces (except within a <defs> block). The id attribute of a <traceFormat> allows the format to be reused by multiple contexts (see the Context section). If no <traceFormat> is specified, the following default format is assumed for all traces:

<traceFormat id="default">
   <regularChannels>
      <channel name="X" type="decimal"/>
      <channel name="Y" type="decimal"/>
   </regularChannels>
</traceFormat>

Thus, in the simplest case, an InkML file may contain nothing but <trace> elements.

3.2 Traces

Contents:

trace ::=
wsp* point+

point ::=
     regularPart intermittentPart?

regularPart ::=
regularValue+

intermittentPart ::=
     ":" wsp* intermittentValue* ";" wsp*

regularValue ::=
qualifier? value wsp*

intermittentValue ::=
value wsp*

value ::=
integer | decimal 

integer ::=
'-'? digit+

decimal ::=
'-'? digit+ "." digit+

digit ::=
     "0" | "1" | "2" | "3" | "4" | "5" | "6" | "7" | "8" | "9"




qualifier ::=
     "!" | "'" | """

wsp ::=
     #x20 | #x9 | #xD | #xA

The <trace> element is used to record the data captured by the digitizer. It contains a sequence of points encoded according to the specification given by the <traceFormat> element.

The type attribute of a <trace> indicates the pen contact state (either "pen-up" or "pen-down") during its recording. A value of "indeterminate" is used if the contact-state is neither pen-up nor pen-down, and may be either unknown or variable within the trace. For example, a signature may be captured as a single indeterminate trace containing both the actual writing and the trajectory of the pen between strokes. A value of "continuation" means both that the pen contact state is retained from the previous trace element and that the points of the current trace element are a temporally contiguous continuation of (and thus should be connected to) the previous trace element. This allows a trace to be spread across several elements for purposes such as streaming.

Regular channels may be reported as explicit values, differences, or second differences: Prefix symbols are used to indicate the interpretation of a value: a preceding exclamation point (!) indicates an explicit value, a single quote (') indicates a single difference, and a double quote prefix (") indicates a second difference. If there is no prefix, then the channel value is interpreted as explicit, difference, or second difference based on the last prefix for the channel. If there is no last prefix, the value is interpreted as explicit.

A second difference encoding must be preceded by a single difference representation; which, in turn, must be preceded with an explicit encoding.

NOTE: All traces must begin with an explicit value, not with a first or second difference. This is true of continuation traces as well. This allows the location and velocity state information to be discarded at the end of each trace, simplifying parser design.

Intermittent channels are always encoded explicitly, and prefixes are not allowed.

Both regular and intermittent channels may be encoded with a wildcard character *. The wildcard character means either that the value of the channel remains at the previous channel value (if explicit), or that the channel continues integrating the previous velocity and acceleration values.

Booleans are encoded as "T" or "F".

For each point in the trace, regular channel values are reported first in the order given by the <traceFormat>. If any intermittent values are reported for the point, the set of intermittent values is preceded by a colon and ended with a semicolon. Within these delimiters, the intermittent channels are represented in the order given by the <traceFormat>. The list may be terminated early with the semicolon, and the unreported intermittent channels are interpreted with wildcards.

Here is an example of a trace of 11 points, using the following traceFormat:

<traceFormat>
   <regularChannels>
      <channel name="X" type="decimal"/>
      <channel name="Y" type="decimal"/>
   </regularChannels>
   <intermittentChannels>
      <channel name="B1" type="boolean" default="F"/>
      <channel name="B2" type="boolean" default="F"/>
   </intermittentChannels>
</traceFormat>

<trace id = "id4525abc">

1125 18432'23'43"7"-8 3-5 7  -3 6 2 6 8 3 6:T; 2 4:*T; 3 6 3-6:FF;
</trace>

The trace is interpreted as follows:

An ink markup generator might also include additional whitespace formatting for clarity. The following trace specification is identical in meaning to the more compact version shown above:

<trace id = "id4525abc">
1125  18432
'23  '43
"7  "-8
3  -5
7  -3
6  2
6  8
3  6  :T;
2  4  :  *T;
3  6
3-6  :F  F;
</trace>

<trace id="t001" start="1072915200000"></trace>

<trace id="t002" timeOffset="16202010">...</trace>

<timestamp id="ts001" time="1073026800000"/>
<timestamp id="ts002" timeOffset="600000" timeRef="ts001"/>
<timestamp id="ts003" timeOffset="4320" timeRef="t001"/>

<trace id="t003" timeOffset="180000" 
timeRef="ts001">...</trace>

<trace id="t004" timeOffset="240000" 
timeRef="t003">...</trace>

<trace id="t005" timeOffset="62000" 
timeRef="*">...</trace>

3.2.5 <traceGroup> element

Attributes:

Contents:

The <traceGroup> element is used to group successive traces which share common characteristics, such as the same <traceFormat>. The brush and context sections describe other contextual values that can be specified for a <traceGroup>. In the following example the two traces enclosed in the <traceGroup> share the same brush (see the Brushes section for a description of brushes).

<traceGroup brushRef="penA">
   <trace>...</trace>
   <trace>...</trace>
</traceGroup>

The use of <traceGroup> is reserved for the containment of traces according to their properties at the time of capture. The element may not be nested, and it is not meant to be a generic grouping mechanism for the semantic labelling of traces. For that purpose, InkML provides the <traceRef> element.

Trace groups are the primary mechanism for assigning <context> to traces in archival ink markup. For additional details about this usage, see the Archival Applications section.

4 Context Elements

A number of device, data format, and coordinate system details comprise the context in which ink is written and recorded. These contextual details need to be captured by the ink markup language in order to fully characterize the recorded ink data.

The <context> element) provides various attributes such as canvas and traceFormatRef by which InkML addresses this need. In addition, the <captureDevice> element describes how InkML allows accurate recording of the hardware characteristics relevant during the capture of the ink traces.

Different pen tips (e.g. eraser vs. writing end) or entirely different pens, physical or virtual, may be used on the same input device. These details are captured using the <brush> element.

The following sections describe the elements which are used to capture the context in which the ink data was recorded.

4.1 Capture Device

One of the important requirements for the ink format is to allow accurate recording of metadata about the hardware that was used to acquire the ink contained in a file. This is accomplished in the <captureDevice> block, which may contain either very basic information, or very detailed information about a number of device characteristics.

Some of these characteristics are already commonly used in digitizer specifications, while others are somewhat more esoteric, but nonetheless potentially very useful. Most digitizer manufacturers do not spec them, and many are not able to measure them. However, these device characteristics influence signal fidelity and impose some limits on how the data can be used. Hopefully by beginning to standardize the recording of these characteristics, we can raise awareness and encourage device manufacturers to take them into consideration.

4.1.1 <captureDevice> element

Attributes:

Contents:

Examples:

<captureDevice id="foo"
                manufacturer="AcmePen"
                model="FooBar 2000 USB"
                sampleRate="100"
                uniform="TRUE"
                latency="50">

   <channelList>
      ...

   </channelList>
</captureDevice>

The <captureDevice> element will allow specification of:

• Manufacturer and model
• Basic sampling rate - samples/sec
• Sampling uniformity: must be designated non-uniform if any pen-down points are skipped or if the sampling is irregular
• Latency: latency of the real-time channel, in msec, from physical action to the API time stamp. This is typically specified at the device level, since all channels often are subject to a common processing and communications latency.
• Channel List

The <captureDevice> block, including <channelList>, will often be specified by reference to a separate xml document, either local or at some remote URI. Ideally, <captureDevice> blocks for common devices will become publicly available.

4.1.2 <channelList> element

Attributes:

Contents:

Examples:

<channelList id="foo">
   <channelDef name="X">
      ...

   </channelDef>
</channelList>

The <channelList> element lists all data channels that the device is capable of reporting. Channels include:

• X coordinate (horizontal pen position, relative or absolute)
• Y coordinate (up/down or vertical pen position, relative or absolute)
• Z coordinate (height of pen above paper/digitizer, relative or absolute)
• Force (pen tip force) [NOTE: this is often referred to as "pressure" by manufacturers]
• Tip switch state (touching, not touching digitizer)
• Side switches and Buttons (for example, bezel buttons, cursor buttons...)
• Tilt angle in X dimension
• Tilt angle in Y dimension
• Pen Azimuth (alternative to tilt)
• Pen Elevation (alternative to tilt)
• Pen Rotation (around the pen axis)

In addition, devices may define their own data channels for the recording of device-specific information.

4.1.3 <channelDef> element

Attributes:

Contents:

Examples:

<channelDef name="S">
   <representation  type="boolean"/>
   <threshold       value="0.1" units="newtons"/>
   <skew          value="5" units="msec"/>
</channelDef>

<channelDef name="X">
   <representation  type="integer"/>
   <range           min="0" max="8191"/>
   <resolution    value="0.1"  units="mm"/>
   <quantization  value="0.01" units="mm"/>
   <noise         value="0.05" units="mm"/>
   <accuracy      value="0.5"  units="mm"/>
   <crossCoupling>
      <bind source="Tx"/>
      <bind source="Ty"/>
      <mapping type="mathml" apply="relative">
         <math>
         ...
         </math>
      </mapping>
   </crossCoupling>
   <skew          value="2" units="msec"/>
   <minBandwidth  value="15.0"/>
   <distortion    value=".001"/>

</channelDef>

For each data channel that a device is capable of reporting, its characteristics are described using a <channelDef> element. Each channel may specify any of the following when known and appropriate:

• Value representation - for example, Boolean, integer, or decimal
• Range - the range of possible values that may be reported
• Threshold - (for binary channels) - e.g. the threshold force at which the tip switch is activated

For continuous channels, like X, Y and Z, and Force, these additional characteristics may be specified:

• Resolution - the scale of the values recorded, expressed as "fraction units", e.g. "1/1000 inch") or "decimal units", e.g. "0.1 mm" or "1 degrees" Note that if decimal values are recorded, the quantization of the data may be smaller than the "resolution"
• Quantization - the unit of smallest change in the reported values. If the value is reported as integer, this is assumed to be the same as the resolution
• Noise - the RMS value of noise typically observed on the channel. This is distinct from accuracy! It is an indication of the difference observed in the data from the device when the same path is traced out multiple times (e.g. by a robot).
• Accuracy - the typical accuracy of the data on the channel (e.g. "0.5 mm", "10 degrees" or "0.1 newton") This is the typical difference between the reported position and the actual position of the pen tip (or tilt ...)
• Cross-coupling - the distortion in the data from one channel due to changes in another channel. For example, the X and Y coordinates in an electromagnetic digitizer are influenced by the tilt of the pen. This would be specified by dX/dTx = ... ??? or max delta X vs. Tx = ... ??? If the influencing channels are also recorded, and the cross-couplings are accurately specified, it may be possible to compensate for the cross-coupling by subtracting the influence, at the expense of higher noise. The cross-coupling is always expressed in the units of the two channels, e.g. if X mm and Tx is in degrees, then cross-coupling is in mm/deg
• Skew - the temporal skew of this channel relative to the basic device latency, if any. For example, some devices actually sample X and Y at different points in time, so one might have a skew of -5 msec, and the other +5 msec.
• Minimum bandwidth (in Hz) - the minimum bandwidth of the channel, in Hz (not samples/sec), i.e., the frequency of input motion up to which the signal is accurate to within 3dB.
• Peak rate - the maximum speed at which the device can accurately track motion
• Dynamic distortion, e.g., how velocity affects position accuracy. This is expressed in inverse seconds, e.g. 0.01 mm / mm / sec. This kind of distortion is often cross channel, but this spec only allows a generic, channel independent specification.

The following sections describe each of the characteristics that a channel can specify.

4.1.4 <representation> element

Attributes:

Contents:

4.1.5 <range> element

Attributes:

Contents:

4.1.6 <threshold> element

Attributes:

Contents:

4.1.7 <resolution> element

Attributes:

Contents:

4.1.8 <quantization> element

Attributes:

Contents:

4.1.9 <noise> element

Attributes:

Contents:

4.1.10 <accuracy> element

Attributes:

Contents:

4.1.11 <crossCoupling> element

Attributes:

Contents:

4.1.12 <skew> element

Attributes:

Contents:

4.1.13 <minBandwidth> element

Attributes:

Contents:

4.1.14 <peakRate> element

Attributes:

Contents:

4.1.15 <distortion> element

Attributes:

Contents:

4.1.16 Error Calculations

This Error Calculations section is informative.

The following are some suggestions for how error estimates might be derived from the basic fidelity information in a spatial channel (x or y):

• Total position error is the sum of {absolute accuracy + velocity*(dynamic distortion) + noise + quantization error} for identical path (in all channels).
• Repeatability is also the sum of {noise + quantization error} for a repeated, identical physical trajectory across the digitizer.
• Relative position error is the minimum of {linearity*delta, absolute accuracy). This effects the ability to accurately measure the length and orientation of a short stroke.
• Maximum error including skew (by assuming that all channels are in sync) is equal to the sum of {absolute accuracy + velocity*dynamic distortion + cross-coupling + velocity*(skew) + noise + quantization error}.

All errors are subject to additional distortion from a signal exceeding the channel bandwidth.

4.2 Brushes

Along with trace data, it is often necessary to record certain attributes of the pen during ink capture. For example, in a notetaking application, it is important to be able to distinguish between traces captured while writing as opposed to those which represent erasures. Because these attributes will often be application specific, this specification does not attempt to enumerate the brush attributes which can be associated with a trace. It also does not provide a language for describing brush attributes, since it is possible to imagine attributes which are described using complex functions parameterized by time, pen-tip force, or other factors. Instead, the specification allows for capturing the fact that a given trace was recorded in a particular brush context, leaving the details of precisely defining specific attributes of that context (such as width and color) to a higher-level, application specific layer.

Depending on the application, brush attributes may change frequently. Accordingly, there should be a concise mechanism to assign the attributes for an individual trace. On the other hand, it is likely that many traces will be recorded using the same sets of attributes; therefore, it should not be necessary to explicitly state the attributes of every trace (again, for reasons of conciseness). Furthermore, it should be possible to define entities which encompass these attribute sets and refer to them rather than listing the entire set each time. Since many attribute sets will be similar to one another, it should also be possible to inherit attributes from a prior set while overriding some of the attributes in the set.

4.2.1 <brush> element

Attributes:

Contents:

In the ink markup, brush attributes are described by the <brush> element. This element allows for the definition of reusable sets of brush attributes which may be associated with traces. For reference purposes, a brush specifies an identifier which can be used to refer to the brush. A brush can inherit the attributes of another <brush> element by including a brushRef attribute which contains the id of the referenced brush. As noted above, the definitions of specific brush attributes such as color and width are left to a higher-level markup layer.

Brush attributes are associated with traces using the brushRef attribute. When it appears as an attribute of an individual <trace>, the brushRef specifies the brush attributes for that trace. When it appears as an attribute of a <traceGroup> element, the brushRef specifies the common brush attributes for all traces enclosed in the <traceGroup>. Within the <traceGroup>, an individual trace may still override the traceGroup's brush attributes using a brushRef attribute.

Brush attributes can also be associated with a context by including the brushRef attribute on a <context> element. Any traces which reference the context using a contextRef attribute are assigned the brush attributes defined by the context. If a trace includes both brushRef and contextRef attributes, the brushRef overrides any brush attributes given by the contextRef.

In streaming ink markup, brushes are assigned to a trace according to the current brush, which can be set using the <context> and <brush> elements. See section Streaming Applications for a detailed description of streaming mode.

4.3 Context

This section describes the <context> element and its attributes: canvas, canvasTransform, traceFormatRef, captureDeviceRef, and brushRef. The context element both defines a useful shared context (canvas) and serves as a convenient agglomeration of contextual attributes. It is used by the <traceGroup> element to define the complete shared context of a group of traces or may be referred to as part of a context change in streaming mode. In either mode, individual attributes may be overridden at time of use. Additionally, individual traces may refer to a previously defined context (again optionally overriding its attributes) to describe a context change that persists only for the duration of that trace.

Although the use of the <context> element and attributes is strongly encouraged, default interpretations are provided so that they are not required in an ink markup file if all trace data is recorded in the same virtual coordinate system, and its relationship to digitizer coordinates is either not needed or unknown.

A shared context, called a canvas, is defined as an attribute of the <context> element so the ink markup can support screen sharing amongst multiple devices, each of which might have a different set of capture characteristics. A canvas is simply a unique string, identifying the shared space. A default canvas is defined as "default", and is sufficient to allow simple single-canvas sharing without further action on the part of devices or applications.

An example use for such a shared context might be a single ink markup stream or file that contains traces captured on a tablet computer, a PDA device, and an opaque graphics tablet attached to a desktop computer. The size of these traces on each capture device and corresponding display might differ, yet it may be necessary to relate these traces to one another. They could represent scribbles on a shared electronic whiteboard, annotations of a common document, or the markings of two players in a distributed tic-tac-toe game.

The trace data for these different ink sessions could be recorded using the same set of virtual coordinates; however, it is often useful, and may even be necessary at times, to record the data in the capture device coordinates, in order to more precisely represent the original capture conditions, for compactness, or to avoid round-off errors that might be associated with the use of a common coordinate system. Thus we define a canvasTransform attribute, which is likely to vary from device to device, to capture the mapping from the trace coordinate system to the shared canvas coordinate system. This trace-to-canvas transform is expressed as a standard 2x3 2D transformation matrix (at this time, we ignore the additional complication of any nonlinearity in the digitizing device's coordinate system). The default transform is the identity matrix (with a zero offset).

The format of the trace data--both the mapping from digitizer to trace coordinates and the channels and channel formats present in the data--may also vary from device to device, including from stylus to stylus with the same tablet. Therefore, the <context> element also contains a traceFormatRef attribute, which refers to a specific <traceFormat> element, and a captureDeviceRef attribute, which refers to the <captureDevice> element for the device.

Finally, the <context> element provides a brushRef attribute to record the attributes of the pen during the capture of the digital ink, for a particular context.

4.3.1 Canvas Math

In order to render data from a participant in a multi-party ink app, it is necessary to know how to transform trace data to screen coordinates.

Each party may have a different coordinate system for their traces. Each party will need a coordinate transform to their display that allows scrolling and zooming. Call this S[k].

Party k still needs to determine the meaning of the traces from party i. This is most simply accomplished by having each party define the relationship between their trace coordinate system, and an arbitrary reference coordinate system.

This virtual coordinate system does not have any physical dimensions, because each party will render it differently, and each person will draw onto it differently, with arbitrary zoom and scrolling. Thus the virtual coordinate system is arbitrary.

This virtual coordinate system is provided by the canvas, declared via the canvas attribute. This uniquely identifies a shared virtual coordinate system for cooperating ink applications. Together with the trace-to-canvas coordinate transform (discussed below), it provides a common frame of reference for ink collected in multiple sessions on different devices. In the example above, trace data collected from the tablet computer can be combined with trace data collected from the PDA by specifying a common canvas and describing the relationships between each device's trace data and the common canvas coordinate system.

In the ink markup, the canvas is an unbounded space oriented so that x and y coordinates increase as one moves to the right and down, respectively. Specifying a standard handedness for the canvas coordinate system allows each device to orient and display ink from every other device.

To collaborate in the multi-party ink exchange, party k needs to know the orientation and handedness of the virtual coordinate system (in order to determine their own local S[k]), plus the transform of each other party's data to that virtual coordinate system. Call these transforms T[i]

To map from trace coordinates to screen coordinates, we compose the transform from party i to virtual space with our transform from virtual space to screen space, S[k]. This is M = T * S. This matrix is used to transform all points from that traceGroup.

When the display is zoomed or scrolled, S[k] changes, and M is recomputed. When a new traceGroup with a different T[i] is encountered, it is composed with S[k], and rendering continues.

The S[k] matrix is not part of the inkML file, but is determined locally during capture or rendering.

T and S are the minimum necessary information to be able to render data. However, in order to determine S or T, it is also necessary to make a decision about the orientation of the virtual space. If everyone makes this determination independently, there is no common virtual space. Consequently, the virtual space, or canvas is defined to have a specific orientation.

The orientation of this canvas does not effect anyone, as it disappears when T and S are composed. It simply provides a common intermediate space that everyone uses when computing T (which goes into the xml) and S (which is used only to display the data).

Note: As it is primarily intended as an input specification, the ink markup language does not provide a mechanism for representing the transformations to screen or view coordinates, which relate to ink display and are typically transient.

4.3.2 <context> element

Attributes:

Contents:

Examples:

<context id="context1" canvas="canvas1"
         traceFormatRef="format1" brushRef="brush1"/>
<context id="context2" contextRef="context1" 
         brushRef="brush2"/>
<context id="context3" canvas="canvas1" 
         canvasTransform="2 0 0 0 2 0"
         traceFormatRef="format2" brushRef="brush3"/>

The <context> element consolidates all salient characteristics of one or more ink traces. It may be specified by declaring all non-default attributes, or by referring to a previously defined context and overriding specific attributes. The element is found either in the defs element or as a child of the ink element in Streaming InkML

The first example above is a hypothetical device #1, using a previously defined format1 and brush1, and indicating that it can share trace data using canvas1. Its trace coordinates are mapped to this shared canvas using the default identity matrix with zero offset.

The second example is the same device #1, using a different brush: brush2.

The third example is a hypothetical device #2, using previously defined format2 and brush3, and sharing trace data with the first device by using the common canvas1. Its trace coordinates require a scale factor of 2 to map to the canvas.

5 Generics

This section describes elements of the ink markup which are applicable to multiple aspects of the ink markup.

5.1 Mappings

The <mapping> element provides a uniform syntax for the various uses of mappings in the ink markup. The element has an id attribute, which allows a particular mapping to be applied in multiple places. When a previously defined mapping is reused, the mapRef attribute is used to refer to the <mapping> element, which might be defined in a <defs> block. Mappings appear in the following different places in InkML:

1. In a <channel> element of a <traceFormat>, the <mapping> element is used to describe the transformation from the values actually produced by the device to the values recorded in the trace data.
2. In a <crossCoupling> element, a mapping can be used to specify the cross-coupling effect of one or multiple channels on another channel.
3. A mapping could be used to describe the transformation from trace data values to a common value space (for non-coordinate data), e.g., for pen force data. This, however, is still under discussion and feedback by the community is preferable.

In a <context> element, the canvasTransform attribute is used to describe the 2D transformation from the trace data X-Y values to the common canvas X-Y values (this only applies to coordinate data). This could be considered a form of mapping, but for ease of use, since it is restricted to a 2D transformation matrix, the uniform mapping syntax is not used here. See section <context> element for more information about the canvasTransform attribute.

InkML supports four types of mappings: unknown, identity, lookup table, and formula, which is specified using a subset of MathML. The mapping type is indicated by the type attribute of a <mapping> element. Note: If no mapping appears for a <channel>, it defaults to "unknown", which is safer than assuming that 'X' is identical to the device's 'X' since some filtering or modifications could have been applied. Furthermore, one can specify whether the results of a mapping expression are absolutely or relatively applied to the current data value. This is done by means of the apply attribute. For lookup table mappings in particular, one can determine how to interpret intermediate mapping values. This is specified using the interpolation attribute.

5.1.1 <mapping> element

Attributes

Contents

Identity mappings are specified using an empty mapping element:

<mapping id="m01" type="identity" />

<channel name="X" type="decimal" units="point" default="0"> 
    <mapping  type="identity"/> 
</channel> 

They are used, for example, to define a <traceFormat> channel that reports the exact data that is recorded by a corresponding device channel, with no filtering or transformation.

A lookup table has the following form:

<mapping id="m02" type="lookup" >
  index_1 value_1
  index_2 value_2
  index_3 value_3
</mapping >

Alternatively, the values may appear in the first column of the table, or the table can have multiple index or value columns. The following example means that X += 10 if E == 45, etc...

<channelDef name="X"...>
    ...
    <crossCoupling>
        <bind target="X" column="1"/>
        <bind source="E" column="2"/>
        <mapping id="m03" type="lookup" apply="relative">
            10    45  
            9     50  
            8     55   
            7     60   
        </mapping>
    </crossCoupling>
    ...
</channelDef> 

The value of the interpolation attribute defines the behavior for indices that don't appear in the table. The following summarizes the behavior of the above table for the various values of interpolation:

X += 10  if 45 <= E < 50,
X += 9   if 50 <= E < 55,
...

X += 10  if E <= 47.5,
X += 9   if 47.5 < E <= 52.5, 
...

X += 10  if E <= 45, 
X += 9  if 56 < E <= 50, 
...

X += 10 if E == 45,
X += 9.8 if E == 46, 
...

Formula mappings are specified using a subset of MathML, as follows:

<mapping id="m04" type="mathml" >
  <math>
   ...
  </math>
</mapping>

<mapping id="m05" type="mathml">
  <math mlns=" http://www.w3.org/1998/Math/MathML ">
    <apply>
      <plus/>  
      <ci>Q</ci>
      <cn>10</cn>
   </apply>
</mapping> 

5.1.2 <bind> element

Attributes

Contents

For an identity mapping, if the source channel has a different name than the channel being defined, this can be specified using a <bind> element with a source attribute. In the following markup, the <traceFormat> channel X contains unmanipulated data from the device's devX channel.

<channel name="X">
  <bind source="devX">
  <mapping type="identity">
</channel>

Within a mapping formula (type="mathml"), the variable names in the formula need to be bound to particular channel names. This is specified using a combination of source and variable attributes for binding inputs of the formula, and target and variable for the output of the formula. This is useful if the same mapping formula is to be reused across multiple channels, like X and Y for example.

<bind type="setvar" target="X" variable="Q" />
<mapping id="m06" type="mathml">
  <math mlns=" http://www.w3.org/1998/Math/MathML ">
    <apply>
      <plus/>  
      <ci>Q</ci>
      <cn>10</cn>
   </apply>
</mapping> 

The example shown above means that the channel X is referred to by the variable name Q in the mapping expression "Q+10".

For a lookup table, each index column must be bound to the channel that provides the input for the lookup operation. This is done with a <bind> element that specifies source and column attributes. Similarly, each value column must be bound to the channel that receives the output of the lookup. Its <bind> element specifies target and column.

The following example indicates assignments of channels to columns. It means that values for the channels Tx and P are used to look up the value of the cross-coupling for channel X in the table given by the mapping below:

<bind target="X" column="1"/>
<bind source="Tx" column="2"/>
<bind source="P" column="3"/>
<mapping id="m07" type="lookup" apply="relative">
  10    45    512
  9     50    400
  8     55    372
  7     60    418
</mapping>

5.2 Definitions

5.2.1 <defs> element

Attributes:

Contents:

The <defs> element is a container which is used to define reusable content. The definitions within a <defs> block can be referenced by other elements using the appropriate syntax. Content within a <defs> has no impact on the interpretation of traces, unless referenced from outside the <defs>. In order to allow them to be referenced, elements within a <defs> block must include an id; attribute. Therefore, an element which is defined inside a <defs> without an id, or that is never referenced, serves no purpose.

The three elements which can be defined inside a <defs> are: <context>, <brush> and <traceFormat>. The attributes which are used to reference these definitions are the associated contextRef, brushRef and traceFormatRef attributes. The following simple example illustrates usage of the <defs> element.

<ink>
   <defs>
     <brush id="redPen"/>
     <brush id="bluePen"/>
     <traceFormat id="normal"/>
     <traceFormat id="noForce"/>
     <context id="context1"
              brushRef="redPen"
              traceFormatRef="normal"/>
     <context id="context2"
              contextRef="context1"
              brushRef="bluePen"/>
   </defs>
   <context contextRef="context2"
            traceFormatRef="noForce"/>
   <context id="context3"/>
</ink>

More details on the usage of the <defs> element are provided in the Archival Applications section.

5.3 Annotations

5.3.1 <desc> element

Attributes

Contents

<ink>
  <desc>Robert's signature</desc>
  <trace>
     130 155 144 159 158 160 170 154 179 143 179 129 166 125
     152 128 140 136 131 149 126 163 124 177 128 190 137 200
     150 208 163 210 178 208 192 201 205 192 214 180
  </trace>
</ink>

The <desc> element provides a mechanism for inserting simple textual descriptions in the ink markup. The text contained in the <desc> may include additional information provided by the user generating InkML, and may be displayed by an InkML consumer rendering a graphical representation of traces, for example.

<ink>
  <desc>Einstein's Handwriting</desc>
  <metadata>
    <rdf:RDF xmlns:rdf = "http://www.w3.org/1999/02/22-rdf-syntax-ns#"
             xmlns:dc = "http://purl.org/dc/elements/1.1/" >
        <rdf:Description about="" 
                      dc:language="en"
                      dc:date="2004-04-11"
                      dc:creator="InkML Maker v0.1"
                      dc:publisher="Famous Handwritings Ltd."/>
    </rdf:RDF>
  </metadata>
  <trace>
     130 155 144 159 158 160 170 154 179 143 179 129 166 125
     152 128 140 136 131 149 126 163 124 177 128 190 137 200
     150 208 163 210 178 208 192 201 205 192 214 180
  </trace>
</ink>

6 Streams and Archives

The ink markup is expected to be utilized in many different scenarios. Ink markup data may be transmitted in substantially real time while exchanging ink messages, or ink documents may be archived for later retrieval or processing.

These examples illustrate two different styles of ink generation and usage. In the former, the markup must facilitate the incremental transmission of a stream of ink data, while in the latter, the markup should provide the structure necessary for operations such as search and interpretation. In order to support both cases, InkML provides archival and streaming modes of usage.

6.1 Archival Applications

In archival usage, contextual elements are defined within a <defs> element and assigned identifiers using the id attribute. References to defined elements are made using the corresponding brushRef, traceFormatRef, and contextRef attributes. The following example:

<defs>
   <brush id="penA"/>
   <brush id="penB"/>
   <traceFormat id="fmt1">
     <regularChannels>
       <channel name="X" type="integer">
       <channel name="Y" type="integer">
       <channel name="Z" type="integer">
     </regularChannles>
   </traceFormat>
   <context id="context1" canvas="canvasA"
            canvasTransform="1 0 0 0 1 0" traceFormatRef="fmt1" 
            brushRef="penA"/>
   <context id="context2" canvas="canvasA"
            canvasTransform="2 0 0 0 2 0" traceFormatRef="fmt1" 
            brushRef="penB"/>
</defs>

defines two brushes ("penA" and "penB"), a traceFormat ("fmt1"), and two contexts ("context1" and "context2") which both refer to the same canvas ("canvasA") and traceFormat ("fmt1"), but with different canvas transforms and brushes. Note the use of the brushRef and traceFormatRef attributes to refer to the previously defined <brush> and <traceFormat>.

Within the scope of a <defs> element, unspecified attributes of a <context> element are assumed to have their default values. This <defs> block:

<defs>
   <brush id="penA">
   <context id="context1" canvas="canvasA" brushRef="penA"/>
</defs>

defines "context1", which is comprised of "canvasA" with the default canvasTransform and traceFormat (the identity mapping and a traceFormat consisting of decimal X-Y coordinate pairs), and "penA".

A <context> element can inherit and override the values of a previously defined context by including a contextRef attribute, so:

<defs>
   <brush id="penA"/>
   <context id="context1" canvas="canvasA"
            canvasTransform="1 0 0 0 1 0"/>
   <context id="context2" contextRef="context1"
            canvasTransform="2 0 0 0 2 0" brushRef="penA"/>
</defs>

defines "context2" which shares the same canvas ("canvasA") and traceFormat (the default format) as "context1", but has a different canvasTransform and brush.

Within archival ink markup, traces can either explicitly specify their context through the use of contextRef and brushRef attributes, or they can have their context provided by an enclosing traceGroup. In the following:

<trace id="t001" contextRef="context1"/>...</trace>
<trace id="t002" brushRef="penA"/>...</trace>
<traceGroup contextRef="context1">
   <trace id="t003">...</trace>
</traceGroup>

traces "t001" and "t003" have the context defined by "context1", while trace "t002" has a context consisting of the default canvas, canvasTransform and traceFormat, and "penA".

Traces within a <traceGroup> element can also override the context or brush specified by the traceGroup. In this example:

<traceGroup contextRef="context1">
   <trace id="t001">...</trace>
   <trace id="t002" brushRef="penA">...</trace>
   <trace id="t003">...</trace>
</traceGroup>

traces "t001" and "t003" have their context specified by "context1" while trace "t002" overrides the default brush of "context1" with "penA".

A trace or traceGroup can both reference a context and override its brush, as in the following:

<trace id="t001" contextRef="context1" brushRef="penA">...</trace>
<traceGroup contextRef="context1" brushRef="penA">
   <trace id="t002">...</trace>
</traceGroup>

which assigns the context specified by "context1" to traces "t001" and "t002", but with "penA" instead of the default brush.

In archival mode, the ink markup processor can straightforwardly determine the context for a given trace by examining only the <defs> blocks within the markup and the enclosing traceGroup for the trace.

6.2 Streaming Applications

In streaming ink markup, changes to trace context are expressed directly using the <brush>, <traceFormat>, and <context> elements. This corresponds to an event-driven model of ink generation, where events which result in contextual changes map directly to elements in the markup.

In the streaming case, the current context consists of the set of canvas, canvasTransform, traceFormat and brush which are associated with subsequent traces in the ink markup. Initially, the current context contains the default canvas, an identity canvasTransform, the default traceFormat, and a brush with no attributes. Each <brush>, <traceFormat>, and <context> element which appears outside of a <defs> element changes the current context accordingly (elements appearing within a <defs> block have no effect on the current context, and behave as described above in the archival section).

The appearance of a <brush> element in the ink markup sets the current brush attributes, leaving all other contextual values the same. Likewise, the appearance of a <traceFormat> element sets the current traceFormat, and the appearance of a <context> element sets the current context.

Outside of a <defs> block, any values which are not specified within a <context> element are taken from the current context. For instance, the <context> element in the following example changes the current brush from "penB" to "penA", leaving the canvas, canvasTransform, and traceFormat unchanged from trace "t001" to trace "t002".

<brush id="penA"/>
<brush id="penB"/>
<trace id="t001">...</trace>
<context brushRef="penA"/>
<trace id="t002">...</trace>

In order to change a contextual value back to its default value, its attribute can be specified with the value "". In the following:

<context canvas="canvasA" brushRef="penA"/>
<trace id="t001">...</trace>
<context canvas="" brushRef=""/>
<trace id="t002">...</trace>

trace "t001" is on "canvasA" and has the brush specified by "penA", while trace "t002" is on the default canvas and has the default brush.

Brushes, traceFormats, and contexts which appear outside of a <defs> block and contain an id attribute both set the current context and define contextual elements which can be reused (as shown above for the brushes "penA" and "penB"). This example:

<context id="context1" canvas="canvasA" canvasTransform="2 0 0 0 2 0"
         traceFormatRef="fmt1" brushRef="penA"/>

defines a context which can be referred to by its identifier "context1". It also sets the current context to the values specified in the <context> element.

A previously defined context is referenced using the contextRef attribute of the <context> element. For example:

<context contextRef="context1"/>

sets the current context to have the values specified by "context1". A <context> element can also override values of a previously defined context by including both a contextRef attribute and canvas, canvasTransform, traceFormatRef or brushRef attributes. The following:

<context contextRef="context1" brushRef="penB"/>

sets the current context to the values specified by "context1", except that the current brush is set to "penB" instead of "penA".

A <context> element which inherits and overrides values from a previous context can itself be reused, so the element:

<context id="context2" contextRef="context1" brushRef="penB"/>

defines "context2" which has the same context values as "context1" except for the brush.

Finally, a <context> element with only an id has the effect of taking a "snapshot" of the current context which can then be reused. The element:

<context id="context3"/>

defines "context3", whose values consist of the current canvas, canvasTransform, traceFormat, and brush at the point where the element occurs (note that since "context3" does not specify any values, the element has no effect on the current context).

An advantage of the streaming style is that it is easier to express overlapping changes to the individual elements of the context. However, determining the context for a particular trace can require more computation from the ink markup processor, since the entire file may need to be scanned from the beginning in order to establish the current context at the point of the <trace> element.

6.3 Archival and Streaming Equivalence

The following examples of archival and streaming ink markup data are equivalent, but they highlight the differences between the two styles:

Archival

<ink>
   ...
   <defs>
     <brush id="penA"/>
     <brush id="penB"/>
     <context id="context1" canvas="canvas1"
              canvasTransform="1 0 0 0 1 0" traceFormatRef="format1"/>
     <context id="context2" contextRef="context1"
              canvasTransform="2 0 50 0 2 50"/>
   </defs>
   <traceGroup contextRef="context1">
     <trace>...</trace>
     ...
   </traceGroup>
   <traceGroup contextRef="context2">
     <trace>...</trace>
     ...
   </traceGroup>
   <traceGroup contextRef="context2" brushRef="penB">
     <trace>...</trace>
     ...
   </traceGroup>
   <traceGroup contextRef="context1" brushRef="penB">
     <trace>...</trace>
     ...
   </traceGroup>
   <traceGroup contextRef="context1" brushRef="penA">
     <trace>...</trace>
     ...
  </traceGroup>
</ink>

Streaming

<ink>
   ...
   <defs>
     <brush id="penA"/>
     <brush id="penB"/>
   </defs>
   <context id="context1" canvas="canvas1"
            canvasTransform="1 0 0 0 1 0" traceFormatRef="format1"/>
   <trace>...</trace>
   ...
   <context id="context2" contextRef="context1"
            canvasTransform="2 0 50 0 2 50"/>
   <trace>...</trace>
   ...
   <context brushRef="penB"/>
   <trace>...</trace>
   ...
   <context contextRef="context1"/>
   <trace>...</trace>
   ...
   <context brushRef="penA"/>
   <trace>...</trace>
   ...
</ink>

In the archival case, the context for each trace is simply determined by the <trace> element, its enclosing traceGroup, and contextual elements defined in the <defs> block, while in the streaming case, the context for a trace can depend on the entire sequence of context changes up to the point of the <trace> element.

However, the streaming case more simply expresses the changes of context involving "penB", "context1", and "penA", whereas the archival case requires the restatement of the unchanged values in the successive traceGroups.

The two styles of ink markup are equally expressive, but impose different requirements on the ink markup processor and generator. The working group is considering the usefulness of additional mechanisms for distinguishing between the two forms, such as separate profiles for archival and streaming ink markup. Tools to translate from streaming to archival style might also be of use to applications which work on stored ink markup.

7 Semantic Labelling of Traces

The <traceRef> element provides the basis for most semantic labelling of groups of traces. It is used to annotate traces or sets of points within traces with properties that provide higher-level information about the trace data, for example to indicate that a particular portion of the data represent mathematical symbols. Most often, the properties will be specific to the application producing or consuming InkML data.

7.1 <traceRef> element

Attributes:

The from and to attributes are used to indicate the first and last points in the trace (or group of traces) to which the annotation applies. It is an error if there a traceRef element contains a from attribute but no href attribute

If the href attribute points to a trace element, then the from (respectively to) attribute contains an integer representing the 0-based index of the first (resp. last) point in the trace referenced by this traceRef element.

If the href attribute points to a traceGroup element, then the from (respectively to) attribute contains a colon-separated pair of integers. The first one represents the 0-based index of the first (resp. last) trace within the trace group, and the second one represents the 0-based index of the first (resp. last) point in the trace whose index is given by the trace index.

Contents:

If the traceRef element does not contain an href attribute then it must contain one or more traceRef subelements. The trace properties indicated by the contentCategory and other attributes on a traceRef then applies to all descendant traceRef elements.

Example:

<traceRef id="" contentCategory="math">
    <traceRef href="#trace1"/>
    <traceRef href="sample.inkml#trace2" from="0" to="17"/>
    <traceRef dc:language="fr">
        <!-- a nested traceRef, which has
         attributes of all parent traceRefs, 
         i.e. the points referred to by this traceRef have both 
         properties:
         contentCategory="math" and dc:language="fr" -->
       ...
    </traceRef>
</traceRef>

7.2 contentCategory attribute

One of the common attributes of <traceRef> will be contentCategory, which describes at a basic level the category of content that the traces represent; e.g., "Text/English", "Drawing", "Math", "Music". Such categories are useful for general data identification purposes, and may be essential for selecting data to train handwriting recognizers in different problem domains.

A number of likely, common categories are suggested below. However, since this attribute:

1. is largely application-specific
2. may take on values that are difficult or impossible to predict
3. may be a conjunction of more than one primitive type (e.g., "Text/English and Graphics")

it is defined as a general-purpose string, to be used as necessary by applications. If, however, the data fits conveniently into one of the following basic categories, it is recommended that the appropriate suggested category (and optional sub-category) be used.

Suggested categories:

• Text/<language>[/<script>][/<sub-category>] (e.g., Text/jpn/Kanji, Text/en/SSN)
• Drawing[/<sub-category>] (e.g., Drawing/Sketch, Drawing/Diagram)
• Math
• Music
• Chemistry[<sub-category>]

The language specification may be made using any of the language identifiers specified in ISO 639, using 2-letter codes, 3-letter codes, or country names. Some text may also require a script specification (such as Kanji, Katakana, or Hiragana) in addition to the language.

For some applications it may be useful to provide additional sub-categories defining the type of the data.

Suggested sub-categories for Text:

• SSN (Social Security Number)
• Phone
• Date
• Time
• Money
• URL

Suggested sub-categories for Drawing:

• Sketch (Not suitable for geometric clean-up)
• Diagram (Suitable for geometric clean-up)