匀速直线运动模糊图像的模糊参数鲁棒识别  

Identification of robust blur parameter for uniform linear motion blurred images

作　　者：赵鹏[1,2] 曹军[2] 韦兴竹[1] 

机构地区：[1]东北林业大学信息与计算机工程学院,黑龙江哈尔滨150040 [2]东北林业大学机械工程博士后流动站,黑龙江哈尔滨150040

出　　处：《光学精密工程》2013年第9期2430-2438,共9页Optics and Precision Engineering

基　　金：教育部中央高校基本科研业务费专项基金资助项目(No.DL12DB06);教育部新世纪优秀人才支持计划专项资金资助项目(No.NCET-12-0809);中国博士后科学基金第5批特别资助项目(No.2012T50318)

摘　　要：针对摄像机成像时经常产生匀速直线运动模糊,导致图像退化的现象,提出了恢复图像涉及的精确辨识运动模糊尺度的方法。由于利用Radon变换可以精确地识别运动模糊方向,通过图像旋转可以将运动模糊的方向旋转到水平轴,因此,只对水平方向的运动模糊图像进行研究。对于不带噪声的运动模糊图像,对其进行Fourier变换转化到频率域,使用BP神经网络检测运动模糊尺度;BP神经网络的输入量为频谱图中央区域的幅度加和。对于带噪声的运动模糊图像,先对其进行双谱变换,再使用BP神经网络检测运动模糊尺度;双谱中每列的最大值为BP神经网络的输入量。最后通过仿真实验验证了本文方法的正确性和有效性。验证结果显示,当噪声图像的信噪比SNR≥23 dB时,本文方法的模糊参数辨识平均误差≤5%,优于传统的运动模糊参数辨识方法。Uniform linear motion blur often occurs in the camera imaging,which results in image deg radation seriously. Therefore, this paper proposes a method to identify the motion blur extent accu- rately. Since a motion blur angle could be identified accurately by Radon transform and ~he motion blur direction could be rotated to the horizontal, only the horizontal motion blurred images are needed to be researched. For a motion blurred image without noises, the Fourier transform is used to trans- form it to frequeney domain, the motion blur length is estimated by the BP neural network and the in- put of BP network is the sum of the amplitudes in the central region of the spectrum. For a blurred image with noises, the bispectrum transform is applied, the motion blur length is estimated by BP neural network and the input of BP network is the maximum value of each column of bispectrum. Thcsimulation experiments show that the proposed method in this paper is correct and efficient. When the SNR is larger or equal to 23 dB proposed method is below 5 %, in noisy images, the mean error of blur parameter identification which is superior to that of conventional schemes.

关 键 词：光学成像 运动模糊 噪声图像 图像识别 双谱 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]