Design And Implementation Of Image Retrieval System Based On Multi Feature Values

In recent years, Image retrieval technology has been a hot topic in the field of multimedia. In order to solve the shortage caused by global feature in image retrieval, local feature was put forward and got continuous development. By studying a content-based image retrieval system that contains feature generation module, index module and query module while supporting a variety of global features and some local features. it has important significance.Firstly,the feature generation module provides a variety of image feature extraction methods, includes a variety of mainstream global feature extraction methods such as color histogram and classic local feature extraction methods such as SIFT and SURF. Users can specify one or more query feature extraction algorithms to get the best search results according to the actual situation. At the same time, the pre-processing method based on contour extraction was adopted in this module to judge the main part of the image. This method can work well in most instances, and the feature points will be focused on the main object.Secondly, the existing high-dimensional data indexing methods and problems were studied. And the problems of using different types of features at the same time were solved by adopting the combination of clustering and full-text index method in the index module. In this method, local feature points were clustered to build histograms to keep consistent with the expression of global features and indexed by using the full-text index method with other features at same time. This method can improve the recognition speed and guarantee the accuracy at the same time.Through statistics and analysis to the index build and query time and the query accuracy on different sized data sets, verify that our system can get more accurate results in a short time by combining different features.