基于特征匹配的快速鲁棒数字稳像

针对手持移动摄像装置拍摄视频序列相邻帧间存在平移、小角度旋转运动，而且易受噪声、光照变化的影响等问题，提出一种基于优化Oriented FAST and rotated BRIEF(ORB)特征匹配的实时鲁棒电子稳像算法。对相邻帧预处理后用Oriented FAST算子检测特征点，再用Rotated BRIEF描述提取的特征点并采用近邻汉明距离匹配特征点对，然后采用级联滤波去除误匹配点对，最后使用迭代最小二乘法(ILSM)拟合模型参量进行运动补偿实现稳像。图像匹配测试和稳像实验结果表明:基于改进的ORB算法的电子稳像方法补偿每一帧的时间均小于0.1 s，定位精度可达亚像素级，能有效补偿帧间平移旋转运动，而且对噪声和光照变化有较强鲁棒性。经稳像处理后，实拍视频质量明显提高，峰值信噪比（PSNR）平均提高了10 db。

Fast robust digital image stabilization based on feature matching

In view of the problems that the translation and small angle rotation motion always exist between adjacent frames when the handheld mobile camera is filming video sequence, and it is easily affected by noise and illumination changes,we put forward a kind of real-time robust digital image algorithm based on optimized oriented features from accelerated segment test (FAST) and rotated binary robust independent elementary features (BRIEF) (ORB) feature matching algorithm.Firstly the adjacent frame images were preprocessed to enhance image clarity and to avoid noise interference;secondly the oriented FAST operator was used to detect feature points and the rotated BRIEF was used to describe feature points, then the neighbor hamming distance was adopted to match the ORB feature point pairs; thirdly the cascaded filter was used to remove the false matching points; finally the iterative least squares method(ILSM) was used to fit model parameters,then the motion compensation was done to achieve digital image stabilization. Standard image matching test and digital image stabilization experimental results show that the run time of compensation for each frame by the electric image stabilization method based on improved ORB algorithm is faster than 0.1 s, the positioning accuracy can reach sub-pixel level, this method can effectively compensate the translation and rotation movement between the adjacent frames, and is not sensitive to noise and illumination changes, has strong robustness. After image stabilization processing,the real scene shooting video quality obviously improves and the peak signal-to-noise ratio (PSNR) increases by an average of 10 db.