Kannada Script Recognitions from Scanned Book Cover Images

摘要

Text extraction from the images plays a vital role in providing valuable information. Text extraction from images is still a challenging area specially extracting text from regional scripts of India like Kannada, Malayalam etc. Most of the times the images contain complex background then the cropping of text becomes even more challenging for extracting features. The input image is a scanned document images of Kannada book cover which is scanned with flatbed scanner of 400dpi resolution. The data sets are created by dividing the original images into number of varied size of blocks. Both spatial and frequency features are extracted for classifying images. This paper aims at recognizing the scanned images block which contains text or not by using multiple feature approach. The classification is analysed using Multilayer perceptron, Kstar and KNN. Experiments are performed on different sets of scanned documents of text cover images. Compare to all the classifiers KNN has given the encouraging results.