WSEAS Transactions on Signal Processing
Print ISSN: 1790-5052, E-ISSN: 2224-3488
Volume 9, 2013
Local Pixel Class Pattern Based on Fuzzy Reasoning for Feature Description
Authors: , ,
Abstract: Local features extracted from images have broad potential in varieties of computer vision applications, such as image retrieval, object recognition and scene recognition. However, many of the existing features are not robust enough due to the existence of illumination changes, which is a common occurrence in real world applications, e.g. shadowing. In this paper, a novel feature descriptor is proposed to designing more robust to illumination changes. The basic principle of the proposed method is based on the observation that although the intensity values may be changed due to illumination changes, the texture structure or pixel class in the corresponding locations still remains unchanged. Specifically, they are achieved by applying Histogram Equalization and Intensity Normalization in pre-process step, and considering overall intensity distribution properties together with local intensity difference information by introducing fuzzy reasoning rules. In order to make our descriptor more discriminative and robust, we also propose a novel gradient-based weighting scheme. Experimental results on the popular Oxford dataset have shown that our proposed descriptor outperforms many state-of-the art methods not only under complex illumination changes, but also under many other image transformations.
Search Articles
Keywords: Local feature descriptor, Illumination invariance, Histogram Equalization, Fuzzy reasoning, Local pixel class pattern