运动与离焦耦合的模糊图像参数辨识方法  

Parameter Identification for Mixed Blur Image with Motion Blur and Defocus Blur

作　　者：刘子伟[1] 许廷发[1] 赵鹏[2] 

机构地区：[1]北京理工大学光电学院光电成像技术与系统教育部重点实验室,北京100081 [2]东北林业大学信息与计算机工程学院,黑龙江哈尔滨150040

出　　处：《北京理工大学学报》2014年第3期327-330,共4页Transactions of Beijing Institute of Technology

基　　金：国家自然科学基金重点资助项目(61027002);国家自然科学基金资助项目(60972100);国家"九七三"计划项目(2009CB72400603);国家教育部新世纪优秀人才支持计划专项资助项目(NCET-12-0809)

摘　　要：针对运动和离焦耦合模糊图像运动模糊尺度、角度和离焦模糊半径的辨识难题,提出了一种运动与离焦耦合的模糊图像参数辨识方法.首先,对模糊图像进行Fourier变换,采用Radon变换辨识运动模糊角度;其次,对模糊图像的频谱图进行滤波,将频谱图中央区域的幅度之和作为BP神经网络的输入量,检测运动模糊尺度时,频谱图按列求和,而检测离焦模糊半径时,频谱图按圆形方向求和.仿真实验表明,不带噪声的耦合模糊图像参数辨识误差在6%以内.For the problem of identifying motion blur length, motion blur angle and defocus blur radius in mixed blur image, a method for parameter identification for mixed blur image with motion blur and defocus blur was proposed. Firstly, the blur image was processed with Fourier transformation, then Radon transformation was adopted to identify motion blur angle. Secondly, filter was adopted for the spectrum, the amplitudes in the central region of the spectrum were summed, and the sum was inputted into the BP neural network. The amplitudes were summed up vertically when estimating motion blur length and were summed up circularly when estimating defocus blur length. The simulation results show that the error of parameter identifications of noise free mixed blur images is smaller than 6%.

关 键 词：运动模糊 离焦模糊 神经网络 RADON变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]