Abstract |
Now days the generation rate of image is increasing rapidly. So it is necessary to develop ways of manipulating (indexing, retrieval) the visual information by its content. This is called Content Based Image Retrieval (CBIR). The content of an image can not be described with words, because the description made by a man is subjective. Never the less this is confirmed by the old saying: "an image is worthy as one thousand words". In order to describe the image content, must be used arithmetic features that will be comparative. In this work texture and color are used for image description. Three ways of texture characterization are presented: Discrete Wavelet Frames, MR-SAR and Gabor filters. For comparing these texture features, experiments are performed on a large number of textured images. These experiments are based on image classifications, according to the above texture features. Also an image segmentation algorithm based on DWF texture features is proposed. This algorithm is partially supervised because the number of different texture contents in the image is needed. Experimental results of the segmentation algorithm are performed on synthetic textured images and on images of physical scenes. As for color features an algorithm of image dominant color extraction is proposed. The extraction is based on HSV and Lab color systems, which are compared. The capabilities of dominant color extraction in image description, and distance metrics for comparing them are presented. Also experiments of image retrieval based on texture and color features are performed, on the same images that where used in texture classification. Finally numerous ways of segment extraction from the image according to texture and color are proposed for content description.
|