| Image matching is a fundamental but important issue on the study of computer vision,and it has many applications in varied areas, such as pattern recognition, medical imageanalysis, image retrieval, etc. The process of achieving the image will be influenced by theweather and some human factors, so the images in the same scene will be translating, rotating,scaling or light changing. It is why image matching was applied to find out the locationrelationship of these images. Several efficient methods of image matching has been broughtup by the scholars around world, but there is still many problems need to be solved, forexample, the multi-exposure image matching under dramatic light changing conditions.The main differences between multi-exposure images is the severely non-linear changesin brightness of the images, for this reason, this thesis put forward a keypoint extractionmethod based on the different brightness layer images. Firstly, we build the differentbrightness layer images for each original multi-exposure image by using the contraststretching function with various parameters. Then, using the SIFT (Scale Invariant FeatureTransform) keypoint extraction method detects the keypoints from the different brightnesslayer images. The result showed that, this method can extract more consistent points.This paper proposes the images matching method based on SIFT and Contrast Contextfor the multi-exposure images. It uses the method based on different brightness layer imagesto detect the keypoints, and then constructs descriptor which combined SIFT gradienthistogram with contrast information. The contrast context is the contrast histogram in theneighborhood area which is center of the keypoint. Finally we use the Euclidean distance asthe similarity measure for the keypoints matching. Results of experiments indicate that ourmethod have a well robustness. Performance of this algorithm is better than classical SIFTmethods, and it can obtain more correct matching points. |
PDF Full Text Request