Crossbreed Thresholding Text extraction Procedure for Images using DWT Domain and SVM Classifier

Abstract

This paper proposes a novel crossbreed technique to vigorously extract the texts in images based on Discreet Wavelet Transform (DWT) and Support Vector Machine (SVM). Images on which experimentation has been done are taken from various book covers, newspapers, magazines and commercial products. Database of proposed technique includes 25 images. In addition to that the proposed technique is robust to language selection of the text that is embedded in an image. Experimental database includes images that contain English, Punjabi as well as Hindi font. The proposed technique can be used in the applications such as; keywordbased searching, document retrieving, database collection in an organized manner etc. The projected work is estimated using ICDAR 2013 competition metrics specification and the performance is good as well as results are promising for 3 languages as well.