改进同余变换不变特征的视频拼接算法

VIDEO STITCHING ALGORITHM BASED ON IMPROVED CONGRUENCE TRANSFORMATION INVARIANT FEATURE

视频拼接的平滑过渡问题,是图像处理的难点,为很好地解决这一问题,提出一种基于Sobel算子改进的同余变换不变特征CIF(Congruence transformation Invariant Feature)的视频拼接算法。在图像配准阶段,采集视频序列的关键帧,首先运用改进的CIF算法提取特征点并匹配,然后基于RANSAC算法,估算并优化空间变换矩阵。在图像融合阶段,采用动态连接缝调整算法与色调正常化算法协同工作,以消除图像重合区域的重影与色差。实验表明,该算法不仅能够实现视频的实时拼接,还能很好解决视频拼接的平滑过渡问题。

Smooth transition of video stitching is a nodus in image processing,in order to better solve this problem,the paper puts forward a video stitching algorithm which is based on CIF（ Congruence transformation Invariant Feature） improved by Sobel operator. In the stage of image registration,we collected the key frames in video sequence,and extracted the features by using improved CIF algorithm firstly and matched them with key frames. Then,based on RANSAC algorithm we estimated and optimised the matrix of space transformation. In the stage of image blending,we used the dynamic adjusting algorithm of connect seam and colour tone normalisation algorithm to work collaboratively so as to eliminate the ghosting problem and colour differences in overlapped area between two parts of the image. Experiments showed that the proposed algorithm can achieve real-time video stitching,and can also resolve the smooth transition of video stitching.