Research On Image Mosaic Algorithm Based On Block Matching And Key Point Matching

Image mosaic is one of the earliest and most widely research problems in image processing and computer vision. With the continue development of computer technology, the needs of high-quality wide range panoramic images are growing more and more quickly in virtual reality, remote sensing image processing and medical image analysis. Image mosaic is an effective way to obtain high-quality panoramic images, so the research of image mosaic has become a hot issue in recent years.Image mosaic is mainly based on the similarity of the pixels in superposition areas of the two adjacent images. Thus, the core of image mosaic is how to accurately obtain the pixel coordinates which have the highest degree of similarity in two images. Now, there are many image mosaic methods, for example, the method based on the area (block), the method based on key point, and the method based on phasic.This paper studies the methods which based on the area and based on key point. First , it introduces the meaning, research history and current research hot direction of the image mosaic. Summarized some of the classic image mosaic algorithms and key technologies, analysised the merits and problems of those methods. Combined with existing classic template matching algorithm and row ratio algorithm, gave out a new algorithm which combines with the plane and line. The new algorithm finds the superposition areas of the images effectively, improves the accuracy of the image mosaic and the computing speed. At the same time, this paper draws on the fidelity standard in image compression, gives out a standard to assess the quality of image mosaic.This paper also studies the method based on SIFT feature, improves the process of creating SIFT descriptor in classic SIFT algorithm, reduces the dimension space of the descriptor, speeds up the computing speed in follow-up steps. In image matching, this paper uses RANSAC algorithm to distil the matching points and then calculates the transformation matrix, makes the coordinate transformation for the image which is to be matched. In image fusion, this paper adopts a gauss weighting transition alogrithm to stitch images smoothly.Finally, this paper analysises the shortcomings of the improving algorithm and the difficult problems in image mosaic, points out the direction for further research.