结合SURF的数字稳像技术

Digital Image Stabilization Based on SURF

为了快速准确地稳定视频图像,提出以加速鲁棒特征(SURF)为基础的数字稳像技术.首先针对SURF不适合实时应用的缺陷,根据实际需要和图像尺寸选择皇后模板抽样或者熵值预判来减少建立特征描述子的时间;其次采用基于向量内积的最近邻和次近邻距离比率的方法确定粗匹配结果,并根据特征点本身性质提出级联滤波算法,进一步去除局部匹配点对;最后采用迭代最小二乘法和仿射参数模型求解全局参数并进行反向补偿,得到稳定的视频图像.实验结果表明,该技术能达到有效稳定视频图像的目的,与原SURF算法相比运算时间有极大地提高.

In order to stabilize the video images quickly and accurately, the digital image stabilization technology based on speed-up robust features is proposed. Firstly, under the actual requirements and the image sizes to select the queen pattern or local entropy values as the pre-selection operation, which can reduce the time of building feature descriptors. Secondly, using the distance ratio of the vector inner product from the closest neighbor to the second-closest neighbor, to determine the coarse match set, and then removing the mismatches by the cascade feature filters. Finally, both the iterative least squares method and affine model are used to solve the optimal global motion parameters and compensate the current image, and then obtain stable video images. The simulation results show that the effectiveness of the stabilization algorithm, and the computation time is greatly improved compared to the traditional SURF algorithm.