From the study I realize that the feature extraction methods for word level recognition does not fit well for character level recognition. Thus the question is raised in my mind, what algorithm may work fine for character level recognition as the images for each characters are very small (e.g. 30px by 30px).
Recent works tends to suggest that deep convolutional neural networks (CNNs) are suitable for this. CNNs wrap the feature extraction and decision making in one optimisation, and tend to do quite well.
To get started with convolutional neural networks, you might want to looking at libraries like Theano (see second link). Other libraries (e.g. Keras; third link) wrap libraries like Theano at a high level that is more accessible. The Theano/Keras examples are on MNIST - a well-known dataset for digit recognition (image size 28x28).
For a list of alternative approaches, the first link below lists benchmark results for a lot of publications. It's a bit out of date now, but you can follow the links on each row to understand each pipelines.
http://yann.lecun.com/exdb/mnist/
http://deeplearning.net/tutorial/lenet.html
https://github.com/fchollet/keras/blob/master/examples/mnist_cnn.py
Hi Niranjan,
please go through the following papers, which gives lot of insights on the topic
http://www.worldscientific.com/doi/abs/10.1142/S0218001491000041
http://dl.acm.org/citation.cfm?id=568220
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=4670002&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D4670002
http://www.academia.edu/3648247/An_Introduction_to_the_Process_of_Optical_Character_Recognition
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing