ISOLATED CHARACTERS EXTRACTION USING DIFFER-ENCE-OF-GAUSSIAN FUNCTION

Abstract

A method to extract isolated characters is proposed by using Difference-of-Gaussian(DOG) func-tion. Isolated characters, especially English alphabet or numerical characters, can be seen everywhere in our daily life. How to extract these characters from a digital image efficiently and robustly has been a popular topic for researchers. The DOG function, similar to Laplacian of Gaussian function, was proven to produce the most stable image features compared to a range of other possi-ble image functions. The method incrementally convolves the input image with different scale Gaussian functions and minimizes the computations in high scale images by means of sub-sampling. The candidates of characters are found by connected components analysis in the DOG image and then filtered by group sizes to ignore the un-matched groups. Finally, the experimental results dem-onstrate the success of isolated characters extraction and robustness against noise and illumination change.