Mosaic Algorithm Research Based On Multiview Obtained Images

Image mosaic technology could provide wide viewing and high resolution images.Comparing with scanning cameras and wide-angle lens,this method is convenient and can provide high image quality.It is widely used in many areas,for example,video stitching, virtual reality, aerospace, video compression and transmission, video retrieval and browsing, and so on.The technology is a new research field, with wide range of research areas. This paper mainly studies the image mosaic technology based on image feature points.Feature extraction is the fist step of the image mosaic based on image feature points. The most commonly used feature extraction algorithm include Moravec operator, SUSAN operator, Harris operator, etc.This paper researches the theory of these algorithms and test the effect of Harris operator.SIFT algorithm is a hot research currently.However,the result of SIFT algorithm show that there are false matching points in feature point matching result. False matching points affect the accuracy of transformation matrix.More seriously, this status can cause failure of image registration. In order to remove false matching points, this paper proposes a removing false matching points algorithm based on feature points. Finally,the proposed algorithm is illustrated by different images. Experiment results show that error matching points in registration image are effectively removed and the complexity of time is significantly reduced comparing with RANSAC algorithm.Video image mosaic is an important applications of image mosaic. The first step of video image mosaic is decoding and extracting the video frames. H.264 is the current internationally accepted video codec standard and intra prediction is an important part of H.264. In order to reduce the computational complexity of intra prediction,this paper propses a method based on known prediction mode and the MAD value of neighbor macroblock.Experiment results show that this algorithm can reduce encoding time effectively.Finally, feature points of the frame map are extracted and Matched.Video image registration is completed according to the transformation matrix between frame map.