中国光学, 2013, 6 (3): 378, 网络出版: 2013-07-01
高精度全景补偿电子稳像
High precision digital image stabilization with full frame compensation
电子稳像 SURF算法 全景补偿 兴趣点位移 峰值信噪比 digital image stabilization SURF algorithm full frame compensation feature displacement Peak Signal to Noise Ratio(PSNR)
摘要
针对摄像机拍摄目标过程中自身的随机抖动造成的视频序列不稳定,以及稳像补偿过程中边缘信息的丢失,提出了基于SURF(Speed-up Robust Feature)算法的全景电子稳像方法。首先,运用SURF算法提取当前帧图像和参考帧图像的兴趣点,将两幅图像的兴趣点进行匹配,建立两帧的对应关系。针对兴趣点数目较少及场景中部分区域特征相似的情况,引入了兴趣点位移一致性抑制策略,改进了RANSAC(RANdom SAmple Consensus)误匹配的剔除算法,使得运动矢量的精确度小于1 pixel。然后,判定参考帧的更新策略,获取平滑的运动变量。最后,进行运动补偿,运用图像镶嵌技术对丢失的边缘区域信息进行全景补偿,得到了高精度的全景稳像结果,实验得到的输出视频峰值信噪比(PSNR)提高了33.1%。
Abstract
To overcome the instability of the video sequences caused by the undesirable shakes of a camera, and to reduce the missing of edge information in the process of compensation, a full-frame video stabilization system based on the Speed-Up Robust Feature(SURF) was proposed. Firstly, the SURF was employed to extract the features in the images of the current and the reference frames and to match the features between the two images, so that the correspondence could be established. As a few features were extracted and the features of some areas in the scene were similar, a method of consistency restrain of the features′ displacement was proposed to ameliorate the RANSAC. The motion vector precision is less than 1 pixel. Secondly, by determining the reference frame update strategy, the smoothed inter-frame global motion vector was obtained. Finally, mosaic was used to implement the motion compensation, and the corresponding pixels of the reference frame were filled with a stabilized frame to compensate the unstable motion and to output a stabilized full frame video. The Peak Signal to Noise Ratio(PSNR) is improved by 33.1 percent.
吴威, 许廷发, 王亚伟, 闫辉, 徐磊. 高精度全景补偿电子稳像[J]. 中国光学, 2013, 6(3): 378. WU Wei, XU Ting-fa, WANG Ya-wei, YAN Hui, XU Lei. High precision digital image stabilization with full frame compensation[J]. Chinese Optics, 2013, 6(3): 378.