【物理】图像复原.pptVIP

LIST LIST LIST LIST LIST LIST LIST LIST LIST LIST LIST LIST 图像复原Image Restoration 2009-Fall LIST -*- 任务描述 降质通道 f(x,y) g(x,y) (原始图像） （实际图像） 图像复原 g(x,y) (估计图像） 技术关键： （1）降质通道建模 （2）复原算法 2009-Fall LIST -*- 图像降质模型 (线性模型) H f(x,y) + n(x,y) g(x,y) 2009-Fall LIST -*- 逆滤波(Inverse Filtering) 不稳定 2009-Fall LIST -*- Wiener滤波 2009-Fall LIST -*- 运动模糊模型 x f(x) x0 uΔt 匀速直线运动： Δx Δy 2009-Fall LIST -*- 匀速直线运动参数估计 Cannon方法 a/2 -a/2 h(x) x H(u) u 1/a Identification of blur parameters from motion blurred images Y. Yitzhaky and N.S. Kopeika 2009-Fall LIST -*- Estimation of a correct point spread function (PSF) f(x,y) : given ideal image g(x,y) : the motion blurred image h(x,y) : linear shift-invariant PSF n(x,y) : random noise Exposure time : about 1/30 s → no extreme changes in motion velocity degradation model: assumption: 2009-Fall LIST -*- ACF (Autocorrelation Function) 互相关： 自相关： f Rf t t 用途： 目标定位 相似性判断 2009-Fall LIST -*- (a) original image (b) horizontal motion blurred image horizontal derivative of (b) image derivative : Δf(i,j) Δf(i,j)[0 degree] = f(i+1,j) – f(i,j) Δf(i,j)[k degree] = f(i’,j’) – f(i,j) 2009-Fall LIST -*- ACF d d 实际图像： 运动模糊轨迹不均匀 运动模糊函数不变 迭加增强模糊函数的贡献 2009-Fall LIST -*- Identification of Motion Direction Motion Blurring → Low pass filter intensity at low frequencies suppressed resolution decreased in motion direction intensity at low frequencies increased derivative → High pass filter intensity at high frequencies increased Derivative in motion direction should suppress more of image intensity than that in other directions motion direction → lower power spectrum of derivative 2009-Fall LIST -*- 2009-Fall LIST -*- 算法实现 差分： f(i,j) f(i+1,j) f(i+1,j+1) f(i’,j’) f(i,j) f(i+1,j) f(i’,j’) k 2009-Fall LIST -*- Summary Motion direction k : lowest total intensity of image derivative derivative in motion direction : Δf Average of ACF of image derivative lines : 2009-Fall LIST -*- Results original image horizontal motion blurred(40dB SNR) horizontal derivative 2009-F