Research On Image Mosaic Based On Binary Feature Points

Image mosaic has been a hot issue in the field of digital image processing in the past few years. By building high-resolution or wide-angle images, a single image can present more information. Therefore, image mosaic is widely used in several important fields such as the creation of electronic map, panorama shoot, the construction of virtual reality scenes and the analysis of medical image, etc.Image mosaic is the process of building high-resolution or wide-angle images by stitching two or more images with overlapping areas. In that case, the final image contains the information of every single image. There are two steps to be taken to achieve image mosaic: Image registration and Image fusion.Image mosaic algorithm based on feature points has been widely studied because of its features: efficiency and robustness. This paper aims at thoroughly studying image mosaic algorithm based on binary feature points. Binary feature point refers to the fact that the described information of feature points during image registration is a binary string. Although this kind of algorithms has extremely high efficiency, the accuracy still needs to be improved in image mosaic problems. The main work of this paper is as follows:① Proposing a new binary feature point extraction algorithm. First of all, the image registration algorithms based on feature points by introducing and analyzing the details of two classic algorithms: ORB and BRISK. After analyzing their strengths and weaknesses, this paper proposes a new algorithm which tries to maintain the excellent efficiency and enhance its accuracy in image mosaic problems at the same time. Finally, an experiment has been designed for testing the algorithm.② Conducting a comparative study among image fusion algorithms that are commonly used in image mosaic. The main purpose of this kind of image fusion algorithms is to make the shadow of overlapping area transit more smoothly after being pieced together. Firstly, the paper introduces algorithm ideas like weighted average fusion, median filtering fusion and Laplace pyramid fusion in detail. Afterwards, an experiment has been designed for comparing the effects of these image fusion algorithms in image mosaic problems.③ Proposing an objective evaluation method for image mosaic. The evaluation of image mosaic mainly depends on the man’s subjective evaluation. The result of subjective evaluation may vary from person to person, which is not good for quantitative analysis and high-volume work. Therefore, the paper puts forward an objective evaluation method. Since whether the final image is desirable still belongs to quality issues, this paper puts more emphasis on features of image mosaic. Different from current common objective image quality evaluation method, the method proposed by this paper reflects problems like dislocation and upheaval of overlapping areas more effectively. After introducing the algorithm idea and detailed process of the method, an experiment has been conducted to verify the feasibility of the evaluation method.