Image Retrieve Research Base On Color Texture Clustering Index

At present, due to the traditional text-based search engines such as Baidu, google can not be accurately search images for user demand, so content-based image retrieval (color, texture, shape) is Developing, and become a hot research field. This paper research for Color quantization extraction, clustering index and texture extraction.(1)Use HSV color space. Propose the81,288dimensional quantitative methods, and compare the experimental result of3quantitative methods. The experimental results show that good retrieval results for the288-dimensional quantitative methods.(2)According to quantization of color, propose two clustering indexing algorithm. The first one based on color combinations. Different combination of quantization colors form different cluster, then the image is divided into different clusters. In image retrieval, different combinations of main colors of retrieve image from different cluster name, and then retrieve the cluster corresponding to cluster name, then calculate the similarity of images in the cluster. The second cluster indexing algorithm based on color percentage range. Firstly, it divided color percentage value [0,1] into n intervals, each interval to form a cluster. Image is divided into different intervals clustering base on the percentage value of main color. In image retrieval, set the retrieval interval of main colors, which base on the percentage of the main colors, and then image retrieval accord to combinations and query interval of main color, finally calculate the similarity. This two cluster algorithms avoid calculating the similarity of all the images of the image database, thus improving the speed of retrieval, image retrieval in large databases become possible.(3)Propose a combination of color and texture retrieval method. Firstly, quantify the HSV color space, extract the main color characteristics; then use the GLCM and double tree complex wavelet transform for texture feature extraction. The method extract the main color features and two texture features, experimental results show that the method has better precision and recall rate, with faster retrieval speed.