Image Retrieval Based On Color Textons Features

Color, Texture and Shape feathers are the basic attribute of image. In content -based image retrieval, feature extracting is of great importance. This paper analysis some classic algorithms about feature extracting, and put forward four new discriptors for features extracting and use to image retrieval. They are texton co-occurrence matrix, color textons discriptors, textons correlograms and mutil textons histogram, respectively.Gray level co-occurrence matrices are a classic texture analysis method. It can express the spatial correlation of neighboring pixels, but it can not express color and shape features. This paper put forward texton co-occurrence matrix to describe image features. We have quantized the original images into 256 colors and computed color gradient from RGB color space, and then defined five special textons types for image analysis, and using co-occurrence matrix to describe image features. Image retrieval experimental results have shown that texton co-occurrence matrix method has the discrimination power of color, texture and shape features, the performances are better than that of gray level co-occurrence matrix and color correlograms method.In perceptually uniform color space, we can measure the perceptual difference between two colors using the Euclidean distance. Thus, the perceptual difference between two colors can be measured in a very similar way to the human perception of colors. This paper put forward color textons descriptors to describe image features based on the attribute of perceptually uniform color space. We have quantized the original images into 64 colors and transform the original images from RGB color space into Lab color space, then defined four special types of textons for image analysis and then calculated the color differences of textons to describe image features. Image retrieval experimental results have shown that color textons descriptors has the discrimination power of color, texture and shape features, the performances are better than that of edge histogram descriptor and LBP method.Color correlograms can combine color and spatial information, but its complexity and features dimension are big, thus this paper put forword textons correlograms to express image features. We have quantized the original images into 64 colors, and then defined four special types of textons for image analysis. Finally we have described the attribute of textons image by correlograms. Image retrieval experimental results have shown that the performances of textons correlograms are better than that of color histogram and color autocorrelograms. Textons correlograms also can combine color and spatial information, its complexity and features dimension are small than that of color autocorrelograms significantly.Multi textons histogram is the improvement vision of texton co-occurence matrix. This method has effective utilized the information of edge orientation and color. We have defined four special types of textons for multi textons analysis. It can express the spatial correlation of edge orientation and color in textons images and use to image retrieval. Multi textons histogram can express not only the attribute of edge contain color imformation, but also the attribute of color contain edge imformation. The Experimental results have shown that Multi textons histogram has the discrimination power of color, texture and shape features, the performances are better than that of edge orientation auto-correlogram and edge histogram descriptor significantly.