Research On Image Retrieval Algorithm Using Color Feature And Implementation On DSP

With the rapid development of internet and technologies of computer and multimedia, information of all kinds of images becomes more and more great. It is very important to organize, manage and utilize this multimedia information, which makes content-based image retrieval(CBIR)becomes one of the most active research focuses of implementation of multimedia. In present time, extracting features from image and similarity match are the key issues in CBIR, and research on CBIR focus on these aspects presently.Image feature can be divided into two parts: low visual content and high semantic content. Hopefully, image retrieval can arrive at semantic level which can not be realized with total intelligence because of the present limited technique. The distance between image low level feature and semantic feature is usually called as semantic gap. Retrieval of image low feature which can not describe image content precisely will not match the semantic content when the users do retrieval. So efficient retrieval of image low level feature is still the key difficulty for image retrieval and should be taken further research. And this paper aims at this.At first, we study the related technique on image retrieval and three key techniques based on color feature are discussed: choice of color model, the way of retrieving feature and similarity measurement. As an important visual information, color is stable, not sensitive to size and direction and is one of the earliest features to describe image. Using color in image retrieval has been the most important method. In the expressions of image color character, color histogram is used much often than others. In traditional color quantization, people choose HSV color space, which is suitable to the visual characteristic of human, is utilized. Taking advantage of human's feeling on color, it quantifies color sector with unequal interval, and gets characteristic vector, regarded as color feature of image.In traditional color quantization, which is that all the color quantized to the same level are treated equally but similarity and continuity at the quantization borderland are ignored leading to error, is discussed in this paper. As to thisproblem, a new improved color quantization is introduced: to the fuzzy color quantization method based on HS plan in HSV space, the color which is at the boundary of Hue or Saturation is treated, considering the influence of quantization borderland. If the pixel is in the neighbour of boundary, its color has fuzzy property. Then, the weight of color to every related neighbour will be got and not quantized single area. The non-quantized pixel is considered to belong to the quantization area.This paper will introduce another image retrieval method which is based on FCS(Fuzzy Color-Spatial)because the color histogram just do statistic about the whole distributing. The position and measure of dispersion of pixel are added into this method, so the position of pixels that are on the boundary of quantization affect the position information. We should take the difference that is between the pixels are in non-boundary and boundary into consideration so that color at the borderline can be quantized and the position distributing and dispersal can be improved and do Stat on the information of blur position.Through the experience result, we compare the method we use and traditional extraction of histogram and image retrieval based on FCS. From the result and datum, the method based on FCS is better than the other two and record and precision are higher. Fuzzy color histogram is better than common one and is higher in precision to prove that the method we use in this paper is closer to the people's visual sense. But fuzzy quantization is added into arithmetic, so this takes more time than the traditional method.The development of network promotes the video and image applications. Excellent image quality and good real-time processing needs the increase of the complexity and operation of image processing algorithm, so are the higher demands for image retrieval system. An embedded system based on DSP has flexible programmability and higher performance and lower cost than generic processors thus gained extensive application. We realized this algorithm in TI'TMS320 C6711 DSK.In this thesis we propose a new approach of color quantization method basedon fuzzy quantization to minimize color quantization error, which is close to human's subjective vision perception. Based on the color quantization method, this paper proposes an image retrieval algorithm based on color-spatial feature. Experimental results show the method has significant high retrieval effectiveness. It deserves certain referential value and practical significance in promoting the development of retrieval techniques of image database.