Performance Optimization Study Of Progressive Image Matching Based On Hash Function

Image matching is a fundamental problem in image processing, computer vision and many other related research topics. It is an important and critical step for image retrieval, image localization, image-based three-dimensional reconstruction, and so on. Over the past, some methods had been proposed and achieved a moderate success. However, they majorly focused on improving the matching speed and matching accuracy is not the main concern of these approaches. Epipolar geometry can be used to constrain the relative position between the matching points of two images, thus it is usually used to remove outliers of image correspondences. Although some researchers had also tried to apply epipolar geometry to constraint image matching, they do not consider the matching efficiency in their algorithms. Hence, how to efficiently and accurately matching image feature points in a large data set is a study with high research value.The main contributions are as follows:we integrate the hashing-based matching strategy with the epipolar geometry to propose a progressive hashing-based image matching approach, hashing-based matching approach can promote matching speed significantly, and epipolar geometry can improve the accuracy of image matching result. The epipolar geometry is progressively estimated to filter out incorrect features. Progressive matching indicates that epipolar geometry cannot be very accurately estimated at the initial stage of matching process but its accuracy will be gradually improved as more correct features are identified. The uncertainty of epipolar geometry is used to define the region of filtering out incorrect features, the region is also called envelope of epipolar lines, the feature points which are outside of the envelope of epipolar lines are rejected as false matches.According to the results of experiments, the proposed method can balance the matching efficiency and accuracy and is comparable to the state-of-art approaches in the performance of matching speed and matching accuracy.