F-divergence based local contrastive descriptor for image classification

Recent studies showed that f-divergence based features have achieved great successes in speech recognition, synthesis and dialect classification. This paper proposes a novel local contrastive descriptor for image classification based on the f-divergence, referred as LCD. It extracts local image feature by computing the contrastive characteristic between image patches. Each patch is described as a discrete probability distribution of its properties (e.g. histogram of the intensity or gradient), and the contrast is measured by computing the f-divergence between different distributions. Then we build the bag-of-visual-words (BoVW) model based on the designed LCD and applied it for the task of image classification. We evaluated the proposed descriptor on the widely-used PASCAL VOC2007 benchmark dataset, and experimental results demonstrate that the LCD can work effectively and practically with reasonable accuracy achieved. In addition, we also showed that the LCD, encoding the local color information, can be used to compensate for the gradient-based features (e.g. SIFT) efficiently, with moderate improvements gained.