如何还原模糊的车牌----维纳反卷积滤波算法

Wiener deconv

opencv学习笔记---维纳反卷积滤波算法演示 ,算法参考[维基百科:https://en.wikipedia.org/wiki/Wiener_deconvolution
本例将使用算法实现模糊车牌的清晰化处理,使用Wiener反卷积算法,将因动态拍摄不清楚的车牌还原出来.

licenseplate_motion.jpg Screen Shot 2018-07-10 at 21.59.10.png wienerdeconv.png

维纳反卷积分滤波,常用图像复原,降噪处理,本例中模糊的车牌是运动造成的,通过反卷积滤波,调整卷积滤波的角度,距离和增加噪音,通过反卷积运算,还原出清楚的图像.

from __future__ import print_function

import numpy as np
import cv2 as cv


def blur_edge(img, d=31):
    h, w  = img.shape[:2]
    img_pad = cv.copyMakeBorder(img, d, d, d, d, cv.BORDER_WRAP)
    img_blur = cv.GaussianBlur(img_pad, (2*d+1, 2*d+1), -1)[d:-d,d:-d]
    y, x = np.indices((h, w))
    dist = np.dstack([x, w-x-1, y, h-y-1]).min(-1)
    w = np.minimum(np.float32(dist)/d, 1.0)
    return img*w + img_blur*(1-w)

def motion_kernel(angle, d, sz=65):
    kern = np.ones((1, d), np.float32)
    c, s = np.cos(angle), np.sin(angle)
    A = np.float32([[c, -s, 0], [s, c, 0]])
    sz2 = sz // 2
    A[:,2] = (sz2, sz2) - np.dot(A[:,:2], ((d-1)*0.5, 0))
    kern = cv.warpAffine(kern, A, (sz, sz), flags=cv.INTER_CUBIC)
    return kern

def defocus_kernel(d, sz=65):
    kern = np.zeros((sz, sz), np.uint8)
    cv.circle(kern, (sz, sz), d, 255, -1, cv.LINE_AA, shift=1)
    kern = np.float32(kern) / 255.0
    return kern


if __name__ == '__main__':

    import sys
    fn = 'data/licenseplate_motion.jpg'
    win = 'deconvolution'

    img = cv.imread(fn, 0)
    if img is None:
        print('Failed to load file:', fn)
        sys.exit(1)

    img = np.float32(img)/255.0
    cv.imshow('input', img)

    img = blur_edge(img)
    IMG = cv.dft(img, flags=cv.DFT_COMPLEX_OUTPUT)

    defocus = '--circle'

    def update(_):
        ang = np.deg2rad( cv.getTrackbarPos('angle', win) )
        d = cv.getTrackbarPos('d', win)
        noise = 10**(-0.1*cv.getTrackbarPos('SNR (db)', win))

        if defocus:
            psf = defocus_kernel(d)
        else:
            psf = motion_kernel(ang, d)
        cv.imshow('psf', psf)

        psf /= psf.sum()
        psf_pad = np.zeros_like(img)
        kh, kw = psf.shape
        psf_pad[:kh, :kw] = psf
        PSF = cv.dft(psf_pad, flags=cv.DFT_COMPLEX_OUTPUT, nonzeroRows = kh)
        PSF2 = (PSF**2).sum(-1)
        iPSF = PSF / (PSF2 + noise)[...,np.newaxis]
        RES = cv.mulSpectrums(IMG, iPSF, 0)
        res = cv.idft(RES, flags=cv.DFT_SCALE | cv.DFT_REAL_OUTPUT )
        res = np.roll(res, -kh//2, 0)
        res = np.roll(res, -kw//2, 1)
        cv.imshow(win, res)

    cv.namedWindow(win)
    cv.namedWindow('psf', 0)
    cv.createTrackbar('angle', win, 135, 180, update)
    cv.createTrackbar('d', win, 22, 50, update)
    cv.createTrackbar('SNR (db)', win, 25, 50, update)
    update(None)

    while True:
        ch = cv.waitKey()
        if ch == 27:
            break
        if ch == ord(' '):
            defocus = not defocus
            update(None)

Screen Shot 2018-07-10 at 21.59.10.png