Fast Scene Text Detection with DWT Based Edge Enhancement

The detection and extraction of text in natural and camera captured images is always a challenging problem. The text in images contain most of the important information and can be used in a variety of applications like translators for tourists, aids for visually impaired people, in information retrieval systems etc. . Implementation of such applications on a general purpose computer can be easier, but every time it is not efficient due to additional constraints on memory and other peripheral devices. The paper discusses an efficient yet simple and fast method to extract text from static images. The speed of Haar discrete wavelet transform is fastest among all wavelets because its coefficients are either 1 or -1. It is one of the reasons that Haar DWT is used to detect edges of candidate text regions. Edges are detected horizontally, vertically and diagonally. A Gaussian filtering is initially applied to obtain a smoothed image. Masking is applied on the image sub-bands obtained by DWT. Then on the IDWT image morphological operations like erosion and dilation are applied to remove the noisy areas along with angle tiltation correction and finally ocr output is obtained.