Image Retrieval Based On Geometry Second Moment And Local Feature

Image retrieval is a process of browsing,searching and retrieval images users wanted from a large database of digital images. Image retrieval is one of the key technology of image processing. It is widely used in a geographical information search system, intellectual property protection, medical diagnosis and other fields.With the advent of the era of Big Data, the growth of digital image information and other media information is explosive. Filled with a lot of digital information in the website, it is difficult to search digital information what you want. At the same time, the environment of image retrieval is changed. More and more image retrieval activity is dynamic. Image retrieval is facing the challenge of not only accuracy but also stability. By analyzing the present situation of the development of image retrieval and image retrieval key technologies, the paper proposed method of multi-feature fusion image retrieval. Through a series of experiments, we can prove the feasibility of the method.the completed works is as follows:1.We proposed and verify a possibility of the using of local feature for image retrieval in a complex environment. Due to the complex environment, the image appears to be a variety of changes, it is difficult to extracting and descripting feature. Local feature description for precision image registration can well adapt to this situation Though the experiments, we proved the robustness of SIFT and SURF feature in illumination changes,image blur and other changes of image and find the possibility of SURF feature being used in image retrieval.2. We proposed a fast image retrieval method based on Geometry Second Moment and SURF feature. First, we use Geometry Second Moment for a rough retrieval and narrow your search, then, we use SURF local feature for accurate retrieval. with less loss of retrieval accuracy, the process of image retrieval becomes rapid. Also, we get desired results in the experiments for trademark retrieval. What’s more, we use this algorithm for the experiment of semantic image retrieval and analyzed the pros and cons.