Abstract: Humans around us can be linked through communication. While people with hearing or visual impairments alone can find a mode to share their opinions with others and recognize them, deaf- blind people face a much more difficult communication job. The appropriate technology can play a decisive role. A deaf-blind person is one with impaired senses of hearing and sight. In this paper, we present an approach to analysis of automatic and accurate text detection and recognition of signs for blind persons. The recognized text codes are output to blind users in speech. The proposed approach embeds multi-resolution and multi-scale edge detection, adaptive searching, colour study, and affine rectification in a hierarchical framework for sign detection, with different accents at each phase to maintain the text in customized sizes, orientations, colour distributions and backgrounds. A fast and effective clipping algorithm is planned to extract Maximally Stable Extremal Regions (MSERs) as character candidates using the policy of minimizing regularized differences. The procedure can significantly improve text detection rate and optical character recognition (OCR) accuracy. Performance of the proposed text localization is quantitatively evaluated on ICDAR-2003 and ICDAR-2011 Robust Reading Datasets.

Keywords: MSER, Multi-Resolution, Optical character recognition