双直线运动共同作用的图像模糊参数分析

针对单幅由高频简谐振动和匀速直线运动同时模糊的图像，提出了运动参数辨识的方法. 研究中组合参数不同的两种运动模型，仿真了图像的模糊效果，并应用各种算法进行参数辨识. 结果表明，当两种运动的方向不同时，使用倒谱算法能够准确辨识运动角度；在运动角度上，当微分图像的自相关函数上出现2个十分贴近的正负极值点，且它们之间的距离小于4~5个像素点时，该运动方式为高频简谐振动，否则为匀速直线运动；当运动方式确认后，二次Radon变换算法有利于高频振动获得准确的模糊长度；图像微分自相关函数算法则更利于匀速直线运动获得准确的模糊长度. 

Discernment of Blur Parameters from Blurred Image Caused by Two Linear Motions

The method of how to find the blur parameters on an image was suggested when blurring factors of high frequency vibration and uniform linear motion happen simultaneously. Blurred images with various composite motion blur parameters were synthesized and analyzed using existing algorithms. It was found that cepstral analysis was capable of discerning the motion angles accurately when the directions of two motions are different. Along the motion angle, if the autocorrelation function of derivative image exists two close extreme points with positive and negative values separately and the space between two peak points is less than 4~5 pixels, the image would be blurred by high frequency motion, otherwise it would be blurred by uniform linear motion. Having determined the motion angle and motion style, the quadratic radon transform is the best way to obtain the blur amplitude for high frequency vibration. As for the uniform linear motion, the autocorrelation of image derivative is the best algorithm to obtain the blur length.