基于特征块角点检测的电子稳像方法

Digital image stabilization algorithm based on corner detection in feature blocks

针对稳像过程中的局部运动干扰,提出了一种基于特征块角点检测的电子稳像算法。该算法采用SUSAN算子,对均匀分布的大尺寸特征块进行检测,并构造特征块及对应检索区域角点集合。通过计算基于角点响应函数的修正Hausdorff距离实现背景局部遮挡下的特征块配准,进而通过运动矢量一致性检测得到有效运动矢量。实验结果表明,该方法对实测抖动视频序列的平均峰值信噪比在稳像后提高了19.83 dB,对局部运动干扰下的抖动图像序列具有较好的稳像性能。

This paper proposed a video stabilization algorithm based on feature block comer detection for suppressing local motion interference. It detected corners of uniform distribution feature blocks by using SUSAN operator. It realized feature block registration under partial occlusion by calculating the similarity of the two comer sets with partial mean Hausdorff dis- tance modified with corner response function. It chose the effective motion vectors by making consistency checking. The expe- rimental results show that the average PSNR is raised 19.83 dB after stabilization. The proposed algorithm can stabilize image secmenee with local motion interference.