Finding URLs in images by text extraction in DCT domain, recognition and matching in dictionary
Antoaneta Popova, Johan Garcia, Nikolay N. Neshov, Ivo Rumenov Draganov, Darko Brodić · International Journal of Reasoning-based Intelligent Systems · 2015
Automatic and fast detection of overlaid URLs and text in images and video with adult content can be an important aid for children protection in the internet. In this paper, we propose a method and associated algorithms for URL and text extraction from images and video frames based on combined processing in the compressed and spatial domains. It includes an improved approach for text detection and extraction in the discrete cosine transform (DCT) domain. It also includes an adaptive local binarisation with background compensation in the spatial domain to create an image appropriate for optical character recognition (OCR). As a final step, text recognition and verification is performed by matching to an URL dictionary. The proposed algorithms are quantitatively and qualitatively tested against other contemporary techniques, and encouraging results are obtained. Experimental tests over 3,000 images have been conducted highlighting the efficiency of the proposed approach.