Research On Image Mosaic Based On Harris Feature And Similar Structure Constraint

Image mosaic is a hotspot in image processing which has beensuccessfully applied in remote sensing, medical image processing, videosurveillance, virtual reality, military and other fields. Meanwhile, therequirement of these applications reacts on the image mosaic andpromotes its development. The demand of the image mosaic applicationsfor accuracy, speed and robustness is becoming much higher. This papercarried intensive research on Harris descriptor and similar structureconstraint, and put forward an efficient mosaic algorithm. The maincontribution of this paper are summarized as follows:In the feature extraction and description, the classical Harris cornerdetector which is invariant to rotation, brightness and contrast changeshas been studied. To solve the problem that the Harris corner detector isnot invariant to image scale, the multi-scale Harris corner detection basedon B-spline was proposed. Single-scale Harris corner detector istransformed to multi-scale by B-spline. Keypoints were described byBRISK descriptor which is a robust and fast binary bit string.In the feature matching, the traditional RANSAC（Random SampleConsensus） was improved by using Hamming distance as similaritymeasure, and kd-tree as the search strategy for the initial matching.Similar structure constraint was used to purify initial matching. Then,RANSAC was utilized to calculate the transformation matrix. Thismethod has eliminated the false matching points and reduced theiteratives of RANSAC, thus improves the precision and speed ofmatching.In terms of image fusion, a method was given for calculating theoverlapping area, and using the weighted method to fuse overlapping area.Due to the color differences caused by different cameras, or imagescaptured by the same camera but different times, the color transferalgorithm was adopted to adjust the color after fusion, to produceconsistent tone of image.