A Color-based Image Retrieval Model

With the rapid development of computers science, the Internet and image processing technology, the volume of digital image is going up fabulously. How to get the desired images quickly and efficiently is increasingly becoming an urgent problem. At present, content-based image retrieval has become a hot area of research. In this paper, color-based image retrieval was studied.The study of color-based image retrieval has made a series of achievements at China and abroad. Tom Boyle had raised an image retrieval model based on popular color. Kankanhalli and Mehtre had raised a color matching model based clustering. Pass etc. used a maximum entropy method to analyse the color space information. Stricker etc. had raised a model based on color moments. Colombo etc. had raised a model based on color coherence vector. Cinque etc., who used color clustering method to calculate the color histogram, had raised a model based on the local area.Many quantitative results can not effectively express the characteristics of the images. Many algorithms do not consider the spatial distribution of colors, resulting in some completely unrelated retrieval results.In view of the above questions, we present a model using uniform-based quantization,connected region feature extraction,sort of rectangle information. In the quantitative stage, we present a uniform-based quantization method that let the quantization level was reduced to fifty-four level. In this way, as less quantitative series, it makes the speed of feature extraction faster. In feature extraction stage, we calculate the regional area of the image. We use the color histogram to make a statistics of the image information. Connected region is better to reflect the local information of the image. The method removed some result images that are not similar with the query image, so the high similar image is front of the search results. The main idea is:the results include a result set and a candidate set. We use the rectangle information to remove the dissimilar image from the result set and use the image in the candidate set instead of it. Experimental results show that the method is effective.