Finding the Correspondence Points in Images of Multi-Views

摘要

There are many approaches being proposed to find the correspondence points between two images. They generally perform when used to find the correspondences between two images of the same object, such as in a video sequence or in a stereo camera. However, they fail if the number of true matches between two images was small compared to all the potential correspondence points found, which could happen if both scenes vary significantly, such as when the objects are not the same but of the same class. In this paper, we proposed an approach which uses both the geometric and appearance features to find the true correspondence points even if the number of true matches was small portion compared to all the correspondence points found. The experimental results show that the proposed method also stabilizes very fast compared to the other approaches in the literature.