Invariant Bangla character recognition using a projection-slice synthetic-discriminant-function-based algorithm

Abstract

The projection-slice theorem has been used to create synthetic-discriminant-function based matchedfilters that are capable of discerning rotation and scale distortions. We propose to use the projection- slice algorithm for invariant Bangla character recognition that is capable of accommodating differenthandwriting and font-style distortions. The projection-slice algorithm is applied to training images inorder to synthesize a composite image to represent each class. The opto-electronic fringe-adjustedjoint transform correlator (JTC) technique is then used to correlate the test images with the compositeimages of each class. This technique is chosen for its superior performance to alternative JTCs andthe feasibility of its implementation in the all-optical domain. Simulation results are furnished toprove the effectiveness of the proposed system.