Research And Implementation Of Image Mosaic Algorithm

At present, the wide-view, high resolution image appears more and more importantly inmany fields. As image mosaic technique can solve the contradiction between the view andresolution, the technique has become an important direction of image processing and has highresearch value, as in computer vision, remote sensing image and video supervisory system.Firstly, the article introduced the research present status, characteristics and applicationsof image mosaic, and made a brief description of the process for image mosaic. Then, imagedistortion rectification and image interpolation technique have been analysed in detailed.Moreover, the article researched with emphasis in the core step of image mosaic, which isimage registration. This article detailed introduction the principle, performance as well as theexistence questions of commonly used image registration algorithms.An automatic registration algorithm based on area blocking was proposed to solve theissue of low precision and poor real-time performance in image registration. The methodsearches similar region by normalized cross-correlation of image energy. Then, the SIFTalgorithm was used to extract image feature points in the overlapped region to acquireaccurate matching. In this way, it reduced the searching in useless areas of the classical SIFTalgorithm, and greatly shorten the computation time. In this foundation, the improvedRANSAC was used to purify the matching points to acquire higher precision matching points.After image registration, several image fusion algorithm are introduced, and an newedge-presevation image fusion method was proposed, which preserves more image edgeinformation in the processing of getting seamless image. Experiments results show that, thealgorithm has a good practical significance, which ensures the accuracy of image matching,decreased the image geometric distortion and reduced complexity of image mosaic algorithm.