Comparative Analysis of the Optimization Techniques of Image Filters Used to Improve Optical Character Recognition on Text Images

摘要

Image filtering is changing the appearance of the image by altering the shades and colors of the pixels. Increasing the contrast as well as adding a variety of texture, tones and special effects to images are some of the results of applying image filters for the purpose of improving the success rate of OCR (Optical Character Recognition) performed on text image. Still the main purpose of filtering is reducing the noise around characters in the image. Within Silicon and Software System Limited (S3Group) Company from Dublin, Ireland, the image filters described in this paper are used to improve a subsequent OCR which is part of the process of testing the menu options displayed on the output video of the set-top box under test. This paper focuses on the optimization techniques and on the analysis of which better suits to be applied to each of the image filters in order to make them perform faster. The optimization that produces the best improvement in terms of execution time for all filters above is done using a byte array to store the components i.e. Blue, Green, Red and Alpha of each pixel. Two other optimizations are described, one using C# predefined ColorMatrix class for improving Contrast filter and the other by computing just once the filter weight (i.e. sum of all template elements) for all pixels in the interior of the image in the case of Sharpen and Blur filters, which both use a template to compute the new value of the pixels.