An Impact of Radon Transforms and Filtering Techniques for Text Localization in Natural Scene Text Images

Abstract

AbstractDetection and localization of text in natural scene images have become a greater in need of image-based indexing, searching of images, and in many application areas. Scene text content conveys vital information. The effectiveness of the text localization depends on the efficiency in locating the text regions. In this paper, a detailed exploration of various filter techniques such as radon transform, conservative filter, Gabor filter, bilateral filter, and context-aware filters with maximally stable extremal regions as have been carried out. Maximally stable extremal regions (MSER) give many false positive rates to localize the text regions in the images, hence we have conducted experiments with radon transform and rule-based approach to remove the false positive rate and improve the precision, recall, and F-measure. Each filter has its own application-specific uses. The efficiency of the filtering techniques is examined on the basis of the results obtained for the text localization of scene images. Radon transform has been shown to achieve the best performance among the selected filtering techniques.KeywordsRadon transformConservative filterGabor filterBilateral filterGaussian filter