| Abstract [eng] |
Currently image stitching methods are widely using in scientific researches, such as remote sensing, medical imaging, computer vision or others. So we know, that in scientific could be only two chooses, we will need methods, which can compute images with perfect quality or work fast with giant data. So we choose to compare two feature detection methods: SIFT and Harris corners. Three feature matching methods: SIFT, correlation and monogenic phase. For image transformation we will use affine and homography transformations. So we found out, that SIFT is most stable feature finding method by quality measurements. But Harris corner method works two times faster. In feature matching same as in feature detection, SIFT works better, but slower than correlation and monogenic phase. In general results we found, that SIFT method by structural content (SC) and mean square error (MSE) is a little bit more accurate than other methods. |
