Image Retrieval Technique Research Based On Color And Texture Features

With the rapid development of the network, image data is increasing rapidly. So it is becoming a focus of current research in how to retrieve the desired images from the mass of the images quickly and accurately.Image retrieval with single character always has some limitations, in traditional, color based histogram retrieval method not only has the shift, scale and rotation invariance properties, but also has strong anti noise ability, but forget to fully take into account of the distribution information contained in image space, according to the retrieval defects, this paper presents a method of image retrieval based on color and texture features, taking into account the existing retrieval methods always ignore the differences between computer vision and human vision, this paper proposes two improved retrieval methods respectively after studying the already existing image retrieval methods based on color and texture features. Specific work and innovation of this paper include the following aspects:1. Systematically summarizes the key technologies of content-based image retrieval, including the feature description and extraction, feature similarity and retrieval performance evaluation. After analyze and compare these methods, chose the fit methods to be used in this paper.2. In order to avoid the relationship between neglect spatial variations in the target image and the eye image ignored attention and the different part, color feature extraction method using the overlapped block. At the same time, the center region of the image is weighted with a large value, mainly focus on the human eye, this block method accords with human visual attention mechanism, making the retrieval results more satisfactory.3.Texture feature extraction, it has proposed a kind of retrieval method based on the DCT (discrete cosine transform) domain,according to the characteristics of DC and AC coefficients, extract and use entropy DC coefficients from the difference vector of the AC coefficients to describe the image content features. In the calculation of the adjacent block coefficient difference, different with the literature of the font scanning methods, considering all the relations among the different parts, this paper proposes a method of "Z" shape to encoding in the scanning process, it can be effectively compression the certain zero area, and after the use of entropy coding, image data is further compressed, so as to improve the retrieval accuracy.4. Proposed a comprehensive retrieval method using color and texture features to achieve image retrieval. The retrieval experiments show that integrated retrieval is more consistent with human visual perception than single feature retrieval.5. The experiments use standard Corel and Brodatz image database. In order to compare the different algorithms, we use the recall and precision. Finally, the experiment achieved single texture or color feature retrieval, and also integrated features retrieval.