| Generic
Gesture and Character Recognition |
|
| |
Our research explores generic features
and classification algorithms for recognition of isolated
gestures and characters. With gestures, a key problem is
to learn models from very small numbers of training samples,
and personalize them for the specific user using implicit
feedback. Active-DTW is an example of a classification
method that we have developed that combines the best of
Active Shape Models and Dynamic Time Warping for matching
shapes.
We have also been active in online recognition
of characters and words in Indic languages and scripts.
A supporting activity is the creation of linguistic resources
to support such research. More information about this thread
below:
|
|
| |
An offshoot of our work in handwriting
recognition is the open source Lipi Toolkit, a collection
of tools and algorithms for building handwriting recognition
engines. The toolkit is being used internally as well, for
example: for gesture recognition for the Gesture Keyboard,
a text input solution for Indic languages.
Our research in this area focuses on:
-
Graphical tools for handwriting
data collection
-
Scripts and graphical tools for
the analysis of recognition accuracy and errors
-
Algorithms for handwritten shape
recognition, build scripts for building engines, and
support for UNIPEN 1.0 and a standard shape recognition
interface.
-
The toolkit focuses on:
- supporting collaborative HWR R&D
in academic and industrial settings, tools for user
interface research.
- supporting commercial HWR development.
-
Promotion of standard ink representations
and interfaces.
-
Promotion of sharing & reuse
of tools, algorithms, code and handwriting datasets.
-
Promotion of product and solution
development.
More information on Lipi Toolkit is available
at: lipitk.sourceforge.net
|
|
| |
Collaborative Inking refers to the investigation
of the use of digital ink (pen input) to enable simple kinds
of multi-party collaboration using handwritten annotations.
This leverages our involvement with W3C
InkML, a device and platform independent representation
for digital ink.
The core research problems in this area have
to do with the design of the interaction, secure and efficient
streaming of ink, and synchronization with other modalities
such as voice and images, across multiple client devices
which may vary widely in capabilities
watch
video 
|
|
| |
InkML is an XML data format and a draft
W3C standard for platform and device-neutral representation
of digital ink data that is input with an electronic pen
or stylus. InkML
Toolkit (InkMLTk) is targeted at providing a suite of
tools for working with InkML documents.
The open source toolkit from HP Labs India
includes InkML processor libraries implementing the W3C
InkML specification and different kind of tools such as
Converters (to and from other ink and image formats), InkML
viewers including browser plug-ins, and InkML applications
such as a graphical editor.
More information on InkML Toolkit is available
at: inkmltk.sourceforge.net
|
|
| |
Coffei (Common Forms Framework for Electronic
Ink) developed by HP Labs India allows handwriting captured
as digital ink on paper forms to be directly converted into
machine-readable format using ICR (Intelligent Character
Recognition) and integrated into databases and backend systems.
Coffei supports a number of ink-enabled
devices ranging from stylus-enabled PDAs and TabletPCs to
Anoto pens and electronic clipboards. The use of W3C InkML
for representing digital ink ensures interoperability across
devices and platforms. Coffei also provides rich support
for managing users, forms, devices, and web services for
forms processing.
watch
video
|
|
| |
Click
here for recent Publications listing.
|
|
|
| |
|
 |
|
Printable version
back
top
