Image Retrieval Method Based On Color Histogram And Contour Extraction

With the rapid development of computer and multimedia, the resources of data is rapidly increasing, especially the information of images. Because of this, Content-based image retrieval(CBIR) emerged. With the extension of images-database, the CBIR is developing rapidly.Color is one of the most intuitive visual characteristics in the features of images, which is used in content-based images retrieval widely. For now, color histogram is one of the most widely representation methods of color features. It is easy to calculate, and it has the characteristic of invariance of translation and rotation. Research shows that color histogram can not distinguish the images which has the similar color component for the spatial distribution.As the other visual characters, contour is not a good characteristic which is used for image retrieval directly. The main reason is that computer can not define and distinguish the shape of objects commendably. And for different things, it must use different methods for judging them.Because of above problems, this paper proposes a method which based on color histogram and contour extraction. The main contributions of this dissertation are summarize as follows:First, contour extraction:we use Canny to extract contour of images, which is one of the most best methods of contour extraction. Through smoothing the image, calculating the gradient's amplitude and direction of image, restraining it of not-maximal value, detecting and connecting the edge with threshold value, we can complete the contour extraction.Second, calculate the average of neighbouring colors:through the contour extraction and image partitioning, we can classify the pixels of image. After that, we use different templates to weight the pixels. Because of this, we can calculate the average of neighbouring colors.Third, we can calculate the three-dimensional color histogram which combine the average of neighbouring colors histogram and color histogram.Fourth, image retrieval:we draw the three-dimensional color histogram using the method of above-mentioned for the image which has been given. Next we use euclidean distance to match the characteristics which in the characteristic repository. Then, find out the images with the similarity more than a certain threshold value, and display them decreasingly.The experimental results show that the image retrieval method based on color histogram and contour extraction is better than using the original methods in majority situations.