Realtime handwritten character recognition IFE Team Project proposal

1
Real-time handwrittencharacter recognition(IFE
Team Project proposal)

• Wojciech Bieniecki, Szymon Grabowskiwbieniec,sgr
  abow_at_kis.p.lodz.pl

Lódz, March 2009
2
What?
Develop a lightweight application for
recognition of handwritten characters (letters
and digits) in real-time.
3
Online handwritten character recognition
We can make use of the information on the
writers character strokes (line segments,
writing speed).
4
Handwritten text may be distorted
5
Preprocessing example
6
Application engine (example)
7
Our goals (requirements)
High accuracy (at least 97-98 correct to be
useful). Robust to character translation,
their rotation and scale, character slant,
random distortions. Real-time recognition, the
recognized character displayed immediately. Set
of output classes is fixed (but perhaps we allow
for answer unknown). Word dictionary support
(we may assume English). User-friendly
interface. The writing device is e.g. a mouse,
tablet/stylus. Recommended platform Java and
NetBeans 6.5.
8
Our goals (requirements), contd
Full character set a-zA-Z0-9. Reduced
character sets a-z only,A-Z only, 0-9
only. Dictionary on/off option.
9
Observations, recommendations...
OCR enginei. data acquisition, ii. data
preprocessing, iii. feature selection, iv.
classification. Most important feature
selection. Classification use any fast
off-the-shelf supervised (i.e., using a
training set) classifier, e.g. SLP or MLP neural
networkor 1-nearest neighbor (1-NN) classifier.
10
Verification
Speed matters. Measurements needed.
Accuracy matters even more.Tests with human
volunteers and with readily available
data(google for NIST handwritten char database).
11
(No Transcript)