| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系 === 100 === Feature point matching is to find out the point correspondences between two images of the same scene or object, and this task is a vital part in many images processing technique, such as image matching, object recognition, and other vision-based application. However, there often exist some different kinds of transformation between two images, and will cause bad matching result. To solve the problem, the best way is to construct local descriptor by extracting robust and invariant local feature from interest region, and that will bring out better matching result. SIFT is the most robust local descriptor and has been widely used in many application, but since it is a high-dimensional local descriptor and is complex on feature extracting, the main disadvantage of SIFT is very time-consuming. In order to construct a local descriptor with efficient computation and good matching performance, we refer to Contrast Context Histogram(CCH) which has good matching performance with fast computation and low-frequency DCT coefficients, as well as it keeps important information of an image, and then we proposed a fast and robust local descriptor using features in combining intensity and transformed domains in our study. In experimental results, we can observe that proposed local descriptor has good matching performance under different kinds of transformation. Compared with other local descriptors with good matching performance, proposed local descriptor is much faster on features extracting and has lower dimension, so it has more potential to be used in real-time applications.
|
|---|
