Content-Based Image Retrieval System Of Cultural Images

In the era of digital information, facing the huge data, retrieval has become an important means to extract the effective information. Image retrieval is a retrieval technology widely used, Baidu, Google has entered the field of image retrieval at the earliest. Early image retrieval by artificial image semantic analysis, given text description, then retrieval by keyword matching. However, as the data becomes more and more huge, increasingly high labor costs, manual labeling method becomes unrealistic and efficiency is not high. So, content based image retrieval was proposed. Content based image retrieval method can eliminate the artificial factors involved in, the computer directly analyzes the characteristics of the image itself, given an image, after feature extraction processing can find similar images. In this paper, the content based image retrieval method is applied to the cultural relics image retrieval system, the proposed SIFT features second-order retrieval algorithm for image is a kind of retrieval method based on image content. SIFT features of the image as the search operator, do image analysis and matching, to achieve the goal of retrieval. The main content this paper studied includes:1. Extracting the SIFT feature of image database, do the feature extraction algorithm for each image to obtain the feature points, it's a 128 dimensional vector that contain the information of space and so on. All the extracted feature points stored in the computer as the text.2. The feature points are clustered by clustering, because of thousands of feature points do not bring efficient retrieval. Therefore, it is necessary to reduce the dimension of the feature points efficiently and cluster the visual feature words. The K-Means algorithm can cluster the feature points and find out their center points by the automatic iterative calculation. Finally, the each image feature in the clustering center point on the projection calculation, the characteristics of a two-dimensional frequency distribution table is built, the table can get the distribution of image features in the visual characteristics of these words. At this point, the preparation of the retrieval algorithm is completed. In order to reduce the build time of frequency table. In this paper, hash function was joined to improve the bag of words method, accelerated the frequency table construction speed.3. In order to improve the retrieval efficiency, second-order retrieval of the first order uses the Euclidean distance as similarity retrieval the images that have similar distribution of feature by word frequency table. Second order according to the two image "corresponding" feature points in the same class of features and the proportion of the total number of image points to achieve the exact image retrieval, improve the accuracy of image retrieval.SIFT features second-order retrieval algorithm for image is applied to the actual cultural relics image retrieval system, this paper carries out a lot of experiments to test the actual retrieval results of the algorithm. Retrieval of a given image and similar image in ten thousand pieces of cultural relics. Experimental results show that the method has good recall, in ten similar images seven images can be retrieved as average, and it has a good effect on the query efficiency.