Recognition of Bangla text from scene images through perspective correction

Abstract

This article proposes a scheme for automatic extraction and recognition of Bangla text from natural scene images. An image, when captured by a digital camera may have perspective distortion. Before extracting text symbols, this distortion is corrected using Homography transform. For text extraction, headlines are detected using morphology. First, the components attached or close to the detected headlines, are separated. Further, by applying certain shape and position based conditions we could distinguish text and non-text. Afterwards, by removing the headline we partition the text into two different zones. For recognition purpose, the local chain code histograms of input character are used as features. Finally, separate Multilayer perceptrons (MLPs) are used to recognize text symbols reside in different zones. The classifiers are trained using about 7500 samples of 53 classes. We tested our algorithm on 100 scene images.