Image Retrieval Based On Blocks And Weights-Color Features

With the development of the computer technology, the video, image and sound have become the main part of the internet information. How to locate the image and video that the users need in such plenty of multimedia resource has become the research focus right now. The searching technology based on the content of the image overcomes the disadvantage of the traditional searching technology of the plain-text image, on account of the users can extract the specific feature (color, texture, shape) of the image firstly, and then retrieve the similar image and video from the database.The feature of the image includes color, texture and shape. This paper mainly focuses on the research of searching technology based on the color of the image. This paper realize the color feature extraction through weighted-block of the image, the experiment shows that this algorithm has better performance in spatial information representation and rotation invariant than the traditional method, in the meantime, we use edge histogram with four Sobel operator to improve the weighted block-color feature algorithm, the experiment shows that the new method has better performance in accuracy than traditional algorithm. This paper then formulates an image searching system which allows users to mark focus, based on the above two algorithms.This paper focus on three aspects:1) The article broaches a new searching algorithm based on weighted block-color feature. Research shows that the histogram cannot present the spatial information of the image well; it also gives the disadvantage of the traditional dividing algorithm which can't assure the rotation invariant of the image, the traditional method don't satisfy the regular visual recognition psychology which focus on main part of the image, the main part is usually in the center of the image. Concerning the disadvantage of the two algorithms, this paper gives a new retrieval algorithm of the image features based on weighted block and color feature. Firstly, convert the color space, and then quantify the color space. A new color quantification grade method is raised in this issue by studying the past methods; Secondly, we use rigid segmentation of the image to increase the spatial information, after then we give every block a serial number after the segmentation, then compute their secondary moment and rank them by ascending order. In that case, after the secondary moment ranking, the similar images matching blocks can maintain the largest similarity; additionally, we extract the color feature of every block, then the color feature vector is multiplied by a matching weight, all the eigenvectors summed up to be the eigenvector of the image. The Euclidean distance function will be used to compute the similarity of two images.2) Weighted block-color feature algorithm based on Sobel operator is improved. The same item may have different colors, but their shapes are the same. Concerning that the users always need to search the image's shape, we improve the weighted block-color feature algorithm on the following aspects. We use the weighed block-color feature algorithm as the color representation of the images, use the edge histogram extracted by Sobel operator as the shape representation, then compute the color feature and shape feature distance between the sample image and the image from database separately, then multiply the two distance by matching weight, then sum the results up to be the new similarity distance.3) Design and implement a simple image retrieval system. The users can give blocks different weights according to their needs, the system's core algorithms are these two algorithms proposed and two traditional algorithms integrated. All the experimental data is from the system.In this paper, we proposed two new algorithms according to the disadvantages of the traditional algorithm. Experiment shows that the new algorithms can get better performance than the traditional ones in spatial information representation and rotation invariant. But, the retrieval accuracy of feature extraction algorithm need to improve and further research in order to get a better retrieval result.