High Dimensional Feature Matching and Retrieval Based on Pre-ranking Algorithm

Image retrieval technology has experienced rapid development in recent years and the image retrieval methods based on gist features and SIFT features are increasingly getting more popular. These image features are high-dimensional features and could potentially be simplified to improve the accuracy of image retrieval. In this paper, we studied the existing methods, moved the calculation of the distance between the image and image database offline and ranked the images offline. During a user search, only a few function values are required to be calculated to find the closest image. Based on the ideology of the algorithm, we have carried out the image retrieval based on gist and SIFT features respectively. The experimental results demonstrate that the speed and accuracy of the image retrieval are improved when the image retrieval is applied to the gist features. The accuracy of the image retrieval is improved by 6.76% when image retrieval is applied to SIFT features. Therefore, the pre-ranking algorithm is an excellent image retrieval algorithm for dealing with high-dimensional features.