Toward Improving Content-Based Image Retrieval Systems by means of Text Detection

Abstract

Recent advances in text detection allow for finding text regions in natural scenes rather accurately. Global features in content based image retrieval, however, typically do not cover such a high level information. While characteristics of text regions may be reflected by texture or color properties, the respective pixels are not treated in a different way. In this contribution we investigate the impact of text detection on content based image retrieval using global features. Detected text regions are preprocessed to allow for different treatment by feature extraction algorithms, and we show that for certain domains this leads to a much higher precision in content based retrieval.