直方圖均衡化

直方圖均衡化是通過調(diào)整圖像的灰階分布，使得在0~255灰階上的分布更加均衡，提高了圖像的對比度，達到改善圖像主觀視覺效果的目的。對比度較低的圖像適合使用直方圖均衡化方法來增強圖像細節(jié)

#encoding:utf-8

#
#彩色圖像直方圖
#
from matplotlib import pyplot as plt
import numpy as np
import cv2

src = cv2.imread(r"C:\Users\admin\Desktop\deblur\blur\blur1.jpg")
shape = src.shape
height = shape[0]
weight = shape[1]
# cv2.namedWindow('Original', cv2.WINDOW_NORMAL)
# cv2.resizeWindow('Original', int(weight/4), int(height/4))
# cv2.imshow("Original",src)
# cv2.waitKey(0)

chans = cv2.split(src)
colors = ("b","g","r")
plt.figure()
plt.title("Flattened Color Histogram")
plt.xlabel("Bins")
plt.ylabel("# of Pixels")

for (chan,color) in zip(chans,colors):
    # images,channels,mask,histSize(num_bin),ranges(pix_ranges)
    hist = cv2.calcHist([chan],[0],None,[256],[0,256])
    plt.plot(hist,color = color)
    plt.xlim([0,256])
# plt.show()
plt.savefig(r'C:\Users\admin\Desktop\deblur\blur\hist.jpg')


#
# 灰度圖像直方圖均衡化
#

image = cv2.imread(r"C:\Users\admin\Desktop\deblur\blur\blur1.jpg",0)#讀取灰度圖像
eq = cv2.equalizeHist(image)#灰度圖像直方圖均衡化
# cv2.namedWindow('Histogram Equalization', cv2.WINDOW_NORMAL)
# cv2.resizeWindow('Histogram Equalization', int(weight/2), int(height/4))
# np.hstack() 在水平方向上平鋪
# cv2.imshow("Histogram Equalization", np.hstack([image, eq]))
# cv2.waitKey(0)

#
# 彩色圖像直方圖均衡化
#
chans[0] = cv2.equalizeHist(chans[0])
chans[1] = cv2.equalizeHist(chans[1])
chans[2] = cv2.equalizeHist(chans[2])
eh = cv2.merge(chans)
cv2.namedWindow('Color Histogram Equalization', cv2.WINDOW_NORMAL)
cv2.resizeWindow('Color Histogram Equalization', int(weight/2), int(height/4))
cv2.imshow("Color Histogram Equalization", np.hstack([src, eh]))
cv2.waitKey(0)

plt.figure()
plt.title("EH Flattened Color Histogram")
plt.xlabel("Bins")
plt.ylabel("# of Pixels")
for (chan,color) in zip(chans,colors):
    # images,channels,mask,histSize(num_bin),ranges(pix_ranges)
    hist = cv2.calcHist([chan],[0],None,[256],[0,256])
    plt.plot(hist,color = color)
    plt.xlim([0,256])
# plt.show()
plt.savefig(r'C:\Users\admin\Desktop\deblur\blur\ehhist.jpg')

拉普拉斯增強

拉普拉斯算子可以增強局部的圖像對比度，但對圖像適應(yīng)性較弱，難以用統(tǒng)一的拉普拉斯卷積核對所有的圖像進行增強

import numpy as np
import cv2
from PIL import Image
from matplotlib import pyplot as plt

ori_gray = cv2.imread(r"C:\Users\admin\Desktop\deblur\blur\blur1.jpg",0)  # 讀取灰度圖像
# ori = np.array(ori)
ori_gray = np.array(ori_gray)
weight = ori_gray.shape[0]
height = ori_gray.shape[1]
# laplation kernel
h = np.array(([0,-1,0],[-1,5,-1],[0,-1,0]), dtype="float32")
filteredImg = cv2.filter2D(ori_gray,-1,h)
cv2.namedWindow('Laplation_Result', cv2.WINDOW_NORMAL)
cv2.resizeWindow('Laplation_Result', int(weight/4), int(height/4))
cv2.imshow("Laplation_Result", np.hstack([ori_gray, filteredImg]))
cv2.waitKey(0)

log變換

對數(shù)曲線在像素值較低的區(qū)域斜率大，在像素值較高的區(qū)域斜率較小，逐像素操作，處理速度較慢

圖片.png

import cv2
import math
import numpy as np


def logTransform(c, img):
    # 3通道RGB
    '''h,w,d = img.shape[0],img.shape[1],img.shape[2]
    new_img = np.zeros((h,w,d))
    for i in range(h):
        for j in range(w):
            for k in range(d):
                new_img[i,j,k] = c*(math.log(1.0+img[i,j,k]))'''

    # 灰度圖專屬
    h, w = img.shape[0], img.shape[1]
    new_img = np.zeros((h, w)).astype(np.uint8)
    for i in range(h):
        for j in range(w):
            new_img[i, j] = c * (math.log(1.0 + img[i, j]))

    new_img = cv2.normalize(new_img, new_img, 0, 255, cv2.NORM_MINMAX)

    return new_img


# 替換為你的圖片路徑
img = cv2.imread(r"C:\Users\admin\Desktop\deblur\blur\blur2.jpg",0)
weight = img.shape[0]
height = img.shape[1]
log_img = logTransform(0.8, img)
cv2.namedWindow('log_Result', cv2.WINDOW_NORMAL)
cv2.resizeWindow('log_Result', int(weight/2), int(height/4))
cv2.imshow("log_Result", np.hstack([img, log_img]))
# cv2.imwrite(r'C:\Users\xxx\Desktop\Fourier spectrum2.jpg', log_img)
cv2.waitKey(0)

伽馬變換

圖片.png

γ值以1為分界，值越小，對圖像低灰度部分的擴展作用就越強，值越大，對圖像高灰度部分的擴展作用就越強

20190203144454775.png

import math
import numpy as np
import cv2

def gammaTranform(c,gamma,image):
    h,w,d = image.shape[0],image.shape[1],image.shape[2]
    new_img = np.zeros((h,w,d),dtype=np.float32)
    for i in range(h):
        for j in range(w):
            new_img[i,j,0] = c*math.pow(image[i, j, 0], gamma)
            new_img[i,j,1] = c*math.pow(image[i, j, 1], gamma)
            new_img[i,j,2] = c*math.pow(image[i, j, 2], gamma)
    cv2.normalize(new_img,new_img,0,255,cv2.NORM_MINMAX)
    new_img = cv2.convertScaleAbs(new_img)

    return new_img

img = cv2.imread(r"C:\Users\admin\Desktop\deblur\blur\blur2.jpg",1)
weight = img.shape[0]
height = img.shape[1]
new_img = gammaTranform(1,2.5,img)
cv2.namedWindow('gama_Result', cv2.WINDOW_NORMAL)
cv2.resizeWindow('gama_Result', int(weight/2), int(height/4))
cv2.imshow('gama_Result', np.hstack([img, new_img]))
# cv2.imwrite(r'C:\Users\xxx\Desktop\gray_2.5.jpg',new_img)
cv2.waitKey(0)

參考原文鏈接：
https://blog.csdn.net/dcrmg/article/details/53677739
https: // blog.csdn.net / shawncheer / article / details / 50812705