Improving the quality of optical character recognition method by joint application of wavelet- and curvelet-transforms and vocabulary search algorithm

Abstract

Optical character recognition is a complex problem, which has no definite solution. There are a lot of approaches and methods to solve this problem. The proposed approach, based on aggregate usage of wavelet-transformation for reducing the feature space and probabilistic neural network for classification, has shown a good quality of recognition. However the proposed approach can be improved with preprocessing and postprocessing algorithms. The algorithm of preprocessing based on adaptive thresholding for curvelet and wavelet transformations is proposed. The numerical experiments are held to determine the most efficient algorithm of preprocessing. The approach based on vocabulary search and dynamic programming is proposed for postprocessing.

Keywords

• thresholding
• wavelet-transform
• curvelet-transform
• dynamic programming
• trie-tree
• vocabulary search