光学 精密工程, 2012, 20 (6): 1389, 网络出版: 2012-06-25
双CMOS成像系统中运动模糊图像的复原
Restoration of motion blurred image in dual CMOS imaging system
运动模糊 运动轨迹估计 模糊核 CMOS传感器 图像复原 motion blur motion path estimation blur kernel CMOS sensor image restoration
摘要
为获取高空间分辨率的清晰图像,设计了一种双CMOS成像系统。该系统的两片CMOS传感器可同时获取相同场景的图像,其中一片CMOS传感器获取高帧率、低空间分辨率的图像序列;另一片CMOS传感器获取低帧率、高空间分辨率的运动模糊图像。首先,通过光流法计算高帧率、低空间分辨率CMOS传感器获取图像序列的全局运动路径,在能量守恒和能量与积分时间成正比2个约束条件下估计运动模糊核初始值,通过贝叶斯准则交替迭代优化运动模糊核。最后,利用TV-L1方法从低帧率、高空间分辨率CMOS传感器获取的模糊图像中快速、有效地恢复出清晰图像。仿真和实验结果表明:有38%以上的仿真图像复原结果误差率小于2,且受噪声影响小,复原图像的振铃小。另外,能有效去除实拍图像的空间移不变运动模糊。
Abstract
To obtain high spatial resolution images, a imaging system with two CMOSs was designed to acquire the same scene images simultaneously.Among them, one CMOS sensor was used to acquire high frame rate, low spatial resolution image sequences, and the other one to acquire low frame rate, high spatial resolution images.Firstly,the global motion path was obtained by computing the high frame rate, low resolution image sequences acquired by the CMOS sensor using optical flow method. Then,under the constraint conditions of energy conservation and the energy to be proportional to integration interval,the initial motion blur kernel was estimated and motion blur kernel was optimized by using alternating iterative method based on Bayesian criterion. Finally, a sharp image was restored quickly and effectively from the low frame rate, high spatial resolution blur image acquired by the CMOS sensor using TV-L1 algorithm. Simulation and experiment results indicate that more than 38% simulation images show their error ratios to be less than 2, and has little affected by noises.Restored images have smaller ringing artifacts,and the space-invariance motion blurred photographs can be deblurred effectively.
范赐恩, 陈曦, 张立国, 张虎, 邓德祥. 双CMOS成像系统中运动模糊图像的复原[J]. 光学 精密工程, 2012, 20(6): 1389. FAN Ci-en, CHEN Xi, ZHANG Li-guo, ZHANG Hu, DENG De-xiang. Restoration of motion blurred image in dual CMOS imaging system[J]. Optics and Precision Engineering, 2012, 20(6): 1389.