A Semi-local Method for Image Retrieval

摘要

The visual content of an image is expressed by global or local features. Global features describe some properties of the image such as color, texture and shape. Local features were successfully used for object category recognition and classification to extract the local information from a set of interest points or regions. In this paper, we propose a semi-local method to extract the features based on the previous features extraction methods. Our technique is called the "Spatial Pyramid Matching: SPM". It works by partitioning the image into increasingly fine sub-regions (or blocs) and computing histograms of global features found inside each bloc. The results obtained by the proposed method are illustrated through some experiments on Wang and Holidays Dataset. The obtained Results show the simplicity and efficiency of our proposal.