Image-based Adaptive Image Stitching Algorithm

Image stitching is an important research area of computer processing. It is a technology that carries on the spatial matching to a series of image which are overlapped partly with each other, and finally generates a seamless and big eyespot image by merging these images. It has wide applications in the fields of photogrammetry, computer version, remote sensing image processing, military and so on.The thesis first introduces the present research status, the applications domains and theory of image stitching technology. Then the thesis discusses the basic steps of image stitching. It also deeply analyzes and summarizes the key technologies of image stitching, include their advantages and disadvantages. The thesis also proposes an improved algorithm based on original algorithms. And then it compares the stability, noise resistance and time cost between the improved algorithm and original algorithm by experiments.The improvement on feature points extraction and feature points matching are the key points of the improved algorithm. With the multi-scale Harris corner detection and B-spline function, the improved corner detection can largely suppress the effect of noise. When it refers to feature point matching, it makes use of the characteristics of features point extraction and main gradient vectors of neighbor pixels. In this way, it can more correctly match the features and remove the false matches. The thesis also propose an algorithm based on the affined alternately algorithm to match the feature points of the images with perspective and scale transformation. But its accuracy of this algorithm is not enough, and the algorithm need further improvement. When finishing the feature points matching, the thesis use RANSAC method to calculate the transformation matrix. By remove the outer corner, it can get the best transformation matrix. Finally, the thesis use backward mapping and bilinear interpolation function to execute interpolation calculation in image merge. And then it use overlapped area linear transition method to build a seamless integrated image.