光学 精密工程, 2014, 22 (1): 204, 网络出版: 2014-02-18
改进Noble算子匹配的电子稳像法
Digital image stabilization based on improved Noble feature matching
电子稳像 全局运动 全局一致性 特征匹配 卡尔曼滤波 digital image stabilization global motion global consistency feature matching Kalman filter
摘要
考虑全局运动估计算法对电子稳像系统准确性和实时性的影响,提出了改进Noble算子匹配的电子稳像算法。该算法通过区域预选、特征匹配、参数求解和运动补偿四步操作稳定抖动视频。首先,根据全局运动一致性的特点,采用子块间的绝对差值提前剔除部分不可靠区域,并改进传统Noble算子的全局阈值部分,根据区域梯度均值自适应调整阈值,检测各保留区域的角点,以保证图像特征的空间均匀性分布; 其次,根据特征点的邻域信息构造梯度方向描述子进行粗匹配运算,并利用最近邻次近邻比率和均值距离准则两步操作优化匹配集合; 然后,结合运动模型求解全局参数; 最后,选择卡尔曼滤波过程完成抖动图像的补偿处理。Matlab仿真结果证明: 该算法能平均提高峰值信噪比2 dB以上,当旋转角度小于5°时性能更为优异,能准确、快速地稳定视频图像。
Abstract
In consideration of the effect of global motion estimation algorithm on the real-time performance and the accuracy of a electron stabilization system, a digital image stabilization algorithm based on the improved Noble feature matching was proposed. The algorithm was composed of four steps: preselecting regions, feature matching, solving parameters and motion compensation. Firstly, the global consistency of the background pixels was used to eliminate the unreliable regions, and the adaptive threshold was added into the traditional Noble feature extraction process to ensure the features uniformly distributed in the retained image regions. Then, the neighborhood gradient information of the feature point was used to build the feature descriptors and to get the rough match sets. At the same time, the ratio of the closest neighbor to second-closest neighbor and the mean space distance criteria were both used to optimize the match sets. Furthermore, the global motion parameters were solved. Finally, the Kalman filter was used to get the true shaky components to compensate the images. Matlab experimental results indicate that the algorithm can increase the peak signal-to-noise ratio more than 2 dB, and it is especially excellent when the rotation angle is less than 5 °. In conclusion, the algorithm can stabilize video images quickly and accurately.
初守艳, 席志红. 改进Noble算子匹配的电子稳像法[J]. 光学 精密工程, 2014, 22(1): 204. CHU Shou-yan, XI Zhi-hong. Digital image stabilization based on improved Noble feature matching[J]. Optics and Precision Engineering, 2014, 22(1): 204.