Research On Contented-Based Image Retrieval For Large Scale Images

With the explosive growth of image resources in Internet, the image retrieval technology become a kind of important means that retrieval target image from large scale image database fast and accurately. At present, the technology is considered having wide application prospects in some fields for example general image retrieval, video object tracking, bad information filtering, copyright protection.Traditional content-based image retrieval technology exist performance bottlenecks in the aspects of real-time image retrieval, and it is difficult to support the large scale image retrieval. This paper orients retrieval application for large scale image. In order to enhance the image retrieval efficiency, the paper mainly studies the key technology of content-based image retrieval in three aspects. The research work is as follows:(1) Aiming at the problems of high time complexity in the process of image matching under the environment of large scale images, this paper puts forward a kind of optimization algorithm based on locality sensitive hashing (LSH). Through the algorithm, the paper create index for the global features and SIFT features in the distributed file system. Experimental analysis show that the algorithm has a lower time complexity of image matching compared with the other image features indexing algorithm.(2) In order to reduce computational complexity of multi-feature integration, a two-stage multi-feature fusion method is proposed to improve image retrieval efficiency. First, the algorithm integrates two kinds of linear kernel function which is from the color and texture characteristics of images to calculate the image similarity. And then the method integration of the SIFT characteristics that is a kind local feature of images for the images from the first step. The results of experiments show that the algorithm can offer superior performance of images matching compared with the traditional algorithm.(3) Aiming at the problem of computing and storage for large scale images and image features data, this paper combines the technology of CBIR and distributed computing technology based on hadoop. According to the above two studies, the paper studies the building of image retrieval for large scale images.This paper developed a prototype system that was running on a distributed operating environment based on Hadoop (six nodes). According to the test, the system verified the effectiveness of the results.