Edge Based Segmentation Approach to Extract Text from Scene Images

Scene text in images is finding lots of application in real life. When text in images are extracted efficiently they can be used for guiding tourists, visually impaired people, license plate detection, providing location information etc. Extracting scene text faces many problems, as images are subjected to lots of deprivation such as noise, uneven lighting, blur etc. In our work, we propose an algorithm to extract and analyze the text in from complex scene images efficiently. In the earlier stage, edges are detected using DWT. Then connected component clustering and AdaBoost classifier are used for localizing the text regions. In the next stage morphological operations and heuristic rules are used for character extraction. Finally using OCR extracted texts are analyzed. The proposed algorithm evaluated on different database images has achieved good results in spite of variations in font type, style, orientations and complex background.