Research On Color-based And Texture-based Image Retrieval Algorithm

With the rapid development of the application of multimedia, there is more and more imageinformation, so Content-based image retrieval becomes a hot research field. The main featuresused for image retrieval are color, texture and shape. This paper researches on color-based andtexture-based image retrieval based on plenty of materials. Meanwhile, in this paper also we dosome experiments on the color-based and texture-based image retrieve algorithms proposed. Themain contribution of this paper as follows:Firstly, we discuss color-based image retrieval technique. In this paper HSV color space isintroduced. Then, in order to reduce the boundary discontinuity nonuniform, fuzzy quantizationis used to quantize the pixels in the image. Finally, this paper introduces the formula to calculatethe distance of images. After comparing the traditional color histogram, the experiment resultsshow that the algorithm based on edge color features and fuzzy quantization have the rotation,translation invariance and spatial information. Comparing the block color histogram, thealgorithm in the paper not only keeps spatial information but also strengthen the rotation andtranslation invariance. To the image library used in the experiment, the algorithm can obtainbetter recall.In texture-based image retrieval field, this paper presents an new algorithm. The image isfirstly processed by sharpening, and then combines Gabor filter and Tamura algorithm. Inexperiment, the image processed by sharpening is effective to extract texture. In this paper wecompare the retrieval effectiveness of Gabor filter, Tamura algorithm with the algorithmpresented in this paper. The experiment results show that the algorithm in this paper can producethe best retrieval result. Meanwhile, texture-based image retrieval algorithm reduces the impactof color.In addition, this paper also combines the color-based feather and texture-based feature, andestablishes an experiment retrieval system. Then some implement details are considered. Theweights of color feature and texture feature are decided by the experiment system. Theexperiment results show that the system can obtain good retrieval result using the weightdecided.