The Image Retrieval Method Based On Dominant Color And Contour Approaching

In information society, the image data has increased exponentially over time. The application of the image data has also been expanded in various fields. How to organize, storage, express and retrieve the image data has become an urgent subject.In order to accurately express the information contained in the image, researchers have proposed Content-based Image Retrieval (CBIR) technology. After nearly two decades of research, the technology has become the key for the image expression and retrieval. Currently, domestic and foreign Content-based Image Retrieval has focused on two aspects:One is how to select the appropriate global features to describe the image content and choose one kind of similarity measurement for images; The other is Region-based Image Retrieval, whose main idea is to extract the object in each image region by image segmentation, then use the local features to describe each region and finally integrate all features.Domestic and foreign researches show that the Color Histogram can't reflect the spatial distribution of colors, and Dominant Colors can only reflect a few main colors and their frequency in the region. Besides, Dominant Colors extraction method based on iteration has the shortcoming of high computational complexity. For the shape features, the current method, which based on segmentation or region, can not extract the global shape features of images well. The existing description method of the global shape features can only describe part of the shape features. It's not detailed and accurate enough.In order to express and describe the color characteristics better, based on Dominant Color Descriptors, which was first proposed by Manjunath B. S. and Jens-Rainer Ohm, we gave a definition of Image's Color Distance and showed its calculation and representation method. Then, we analyzed how to use the clustering method to extract Dominant Colors of images, and proposed a Dominant Colors extraction algorithm based on the Non-parametric clustering algorithm. The algorithm combines the advantages of Color Histogram and Color Clustering, because it expresses not only the frequency of the color but also their distribution. Finally, we introduced how to calculate the similarity of Dominant Colors by using image's color distance.In the aspect of the shape characteristics description, we used canny edge detector to extract the image boundary. In order to accurately describe the shapes of the objects in the images, we propose the curve obtaining algorithm and the curve joining algorithm based on the set of boundary points. The algorithm first removed the isolated points in the set of boundary points, and then modified the Curve formed by the points set. Finally, we joined the curves in accordance with the image shape as far as possible. So, all the points became closed contours. This paper used Xiangyang Li's ellipse approaching method to describe the contour characteristics of the objects in the imagesFinally, combining similarity measurement of the color and shape, this paper described how to calculate the similarity between images. Based on this, we proposed Image Retrieval Model based on Dominant Colors and Contour Approaching, then designed and implemented a simulation system to verify these methods.The experimental results show that the image color distance can describe the similarity of images well. The retrieval results are better than using the original method when the colors of the images are rich. Retrieval based on shape characteristics wells up the deficiencies of the retrieval based on color characteristics. Combination of the two retrieval methods is a better way of image retrieval.