Enhanced active contour method for locating text

Abstract

Document analysis of images photographed by camera-equipped mobile phones is a growing challenge. These photos are often poor-quality compound images, composed of various objects and text, this makes automatic analysis complicated, thereby limiting the usefulness of the images. Existing image processing techniques do not manage to clearly decipher the text in such pictures. We developed a method for precisely locating text in complex scene images that can then be further processed by OCR systems. A text kernel operator roughly locates the text in an image. This information then serves as an initialization for the active contour method. This technique enhances the convergence process of the active contour and significantly speeds up the overall process. Moreover, our initialization settings enable systems to easily distinguish between the inside and outside parts of contours. Our experimental results show a significant improvement in the ability to locate and preprocess text. © 2011 IEEE.