Shape Representation and Recognition of Bangla Characters using Fourier Descriptor

Fourier descriptors are used for representing the shapes as well as extracting the features of the characters. It has the advantage of being invariant to the affine transforms. It possesses the ability to reconstruct the original shape. Moreover, all the descriptors are not required to describe the character. Hence huge reduction in feature size is possible. K-nearest neighbor classifier is used to recognize the characters and the accuracy is 98.4% for training set while it is 89.3% for a test set. A case wise in-depth analysis is provided for each misclassified character at the end of the result. General Terms Feature extraction, Fourier descriptor, Bangla character recognition, k-nearest neighbor classifier.