Fast Panorama Stitching And Reducing The Distortion Errors In Image Sequence

The image stitching technology is the important direction in the researching field of digital image processing. It is the basic technology for other related engineering and applications, that aimed at stitching the image lists with overlapping areas in the same coordinate space. As the result of image stitching technology, the thesis not only can form a high-reslution panorama image without stitching seam, but can get the wide view of the scene.The thesis studied the technology of the image stitching based on SIFT(Scale Invariant Feature Transform)features, and focused on improving the efficiency of the stitching process and making the stitching result better. The traditional L-M(Levenberg-Marquardt)algorithm was improved to increase the accuracy of the parameters of the transformation matrix. On the other hand, the thesis put forward improving the traditional image stitching process algorithm. The improved image stitching process taked the middle image of the image sequence as the reference image and then selected the next new input image based on the number of matching points, realizing image stitching process from middle dynamically expanding to both sides. Through the above implements, the number of matching feature points were increased and the quality of panorama result was improved. Finally, the thesis designed and accomplished the image stitching system for image sequence, obtained the panorama result fast. Through a series of experiments, the thesis had proved the practicability and effectiveness of the system. The main research works include:1. In order to eliminate the mismatching points effects on model parameter estimation, the thesis improved the traditional L-M bundle adjustment algorithm. The improved L-M algorithm proposed the parameter optimized based on dynamic weight value. The Huber function was selected instead of reciduals of the standard error function. The weight factor of the Huber function was changed dynamically according to the Euclidean distance between the matching points, that eliminating the influence of mismatching points. Therefore, the improved algorithm increased the accuracy of the parameters of transformation model.2. For the panorama results obtained by the traditional image stitching algorithm exist the distortion phenomenon, the thesis improved the image stitching processing algorithm. The thesis put forward to take the middle image as the reference image, and stitching the images from middle to two sides dynamically. Also, the thesis proposed the new model to get the transformation matrix that map into reference image from anywhere location.3. The thesis researched the Laplace fusion algorithm, solving the problem of stitching seam in overlapping area and adjusting the image light to make the even result in panoramic image. The panorama after Laplace fusion algorithm will be more in line with the requirements of the human eye.4. The thesis designed and accomplished the image stitching system. The system implemented the function that select the appropriate algorithm from control area to stitching image sequence from the input window. Using the system can obtained the high-resolution and high-quality panorama result.Finally, the thesis summarized the research works, and pointed out the future direction for make the panorama result better.