Agent-based automated image descriptor approach for visually impaired people

Abstract

Now-a-days, with the development of high quality graphical software, almost every presentation contains some kind of images. This makes a problem for special users like visually impaired people to understand the presentation, as there is a support for text to voice conversion but there is no a support for images. For mixed documents which contain both pictures and text we propose a hybrid model to make a descriptive text for statistical and geometrical pictures that are easily recognizable. We also use similarity searching with standard annotated templates for both geometrical and non geometrical images to get a meaningful description. At first, the document is categorized according to the associated texts by a neural classifier which is trained using advanced text mining concepts. Then, the similarity matching is performed only in that specific domain and thus, improved matching rate is obtained. Another benefit of the proposed model is that specific matching techniques suitable for a particular domain can be developed and easily adopted to the system. The core of this process is to differentiate geometrical images from ordinary images and we have made a classifier by using these six features information. Out of these six features we have devised four new features based on color projection that can be used for future research in this domain.