A Research Of Image Retrieval Based On Improved Clustering Algorithm

Current image retrieval mainly refers to the content-based image retrieval technology, referred to CBIR technology. Common image features including color, texture feature, shape feature and the semantic features of the image. Each image content features are highdimensional, for large-scale image database, these characteristic data are massive. To reduce the amount of calculation in the retrieval process, speed up the retrieval speed and improve the retrieval accuracy, we need to establish a suitable method for pre-processing the image data library.In this paper, a novel clustering algorithm is used to divide image library in pretreatment. The algorithm is different to traditional clustering algorithms, which is a density-based clustering algorithm for data distribution in any shape, with good adaptability to automatically obtain the cluster center and number. The experimental results show that the proposed clustering algorithm has a shorter time compared to sequential retrieval, which speeds up retrieval of the image effectively. In order to improve the accuracy of image retrieval, this paper introduces the Haar characteristic expression of the texture. The algorithm is a calculation method based on image gray characteristic for a stable image texture features. In this paper, the algorithm for image feature extraction, experimental results show that the feature can effectively improve retrieval precision and recall.In the end, this paper analyzes the advantages and disadvantages of improved clustering and Haar features during the experiment, the experimental results of all the work summarized. Also pointed out the obstacles clustering algorithm and texture feature encountered in the practical application of the current difficulties faced by the image search technology is analyzed, and make a prospect of the next job.