针对现有模糊方向鉴别算法检测小尺度运动模糊图像存在误差的问题, 基于运动模糊图像二次傅里叶变换后的频谱特性, 分析了其误差成因, 并提出了一种适用于小尺度运动模糊图像模糊方向鉴别的新算法。首先对小尺度运动模糊图像的频谱进行细化, 通过中值滤波对细化图像“去噪”, 避开了频谱“噪声”带来的估计误差问题; 其次, 选取灰度中值作为阈值进行二值化, 同时去除二值图像的中心十字亮线和中心亮斑, 使模糊方向特征更加明显; 然后用投影变换计算出亮条纹方向得到小尺度模糊图像的方向估计。最后通过仿真验证了算法在运动模糊尺度为3～10像素的小尺度情况下的实效性。实验结果表明, 该算法相比其他同类算法对于小尺度运动模糊图像的模糊方向估计精度更高。

Aiming at the problem of the blurred direction detection of small-scale motion blurred image, a new algorithm for blurred direction detection was presented by analyzing the cause of detection error. Firstly, we refined the bright lines with median filter in the twice Fourier spectrum to eliminate the spectrum noise. Secondly, we chosen mid-value as a threshold to transform the spectrum to a binary image. And then we removed the cross bright line and bright speck to heighten the direction feature. Finally, we projected all bright spots onto the straight lines of all slopes through the center of the binary image, and calculated the maximum of projection to estimate the blurred direction. Furthermore,the effectiveness of the algorithm with the blue scale of degraded image varied from 3 pixels to 10 pixels was verified by simulation. Experiments demonstrate that the estimation accuracy of this method is higher than other methods in small-scale motion blurred image.