OpenCV对低照度图像的增强

有时候许多图像的对比度较低，在进行特征区域的识别时较为困难，下面我们利用几种方法进行分析，如何提高图像的对比度。
step 1:
利用下面代码，看一下原图像

#导包
import cv2
import numpy as np
import matplotlib.pyplot as plt

#读取外部图像并显示
def ReadAndShowImage():
    image = cv2.imread("00.jpg")
    cv2.imshow("OriginImage", image)
    cv2.waitKey(0)

原图整体较黑，无法清晰明显的得到图像中的数字

step 2:
利用灰度直方图大致看一下灰度分布

#计算图像的灰度直方图
def GainHist():
    image = cv2.imread("00.jpg")
    grayImage = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)
    restHist = cv2.calcHist(grayImage,[0],None,[256],[0,256])
    plt.title("Hist of GrayImage")
    plt.xlabel("Amount of Pixel")
    plt.ylabel("Gray Scale")
    plt.plot(restHist)
    plt.savefig("hist")
    #plt.show()

灰度直方图如下

可以发现，大部分的图像像素都集中在灰度值较小的范围，这也是造成图像较黑的主要原因。能不能将他“拉直”一点，将像素集中到x轴的中间靠上部分
下面代码主要实现这个功能

#直方图均衡化
def EqualHist():
    image = cv2.imread("00.jpg")
    grayImage = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    equalHist = cv2.equalizeHist(grayImage,None)
    hist = cv2.calcHist(equalHist,[0],None,[256],[0,255])
    plt.title("EqualHist of GrayImage")
    plt.xlabel("Amount of Pixel")
    plt.ylabel("Gray Scale")
    plt.plot(hist)
    #plt.savefig("Equalhist")
    # plt.show()
    cv2.imshow("EqualHist",equalHist)
    cv2.imwrite("equalHist.jpg",equalHist)
    cv2.waitKey(0)

上面代码的效果如下图所示


上面经过均衡化后的图像，对比度明显较原图得到增强，效果也比较明显，最后我们要的肯定不是灰度图像，干扰太多了，二值化图像就可以避免这个问题，我们继续向下看

#在对比度受限的情况下，可以使用下面的方法进行
#自适应直方图均衡
def AdaptHist():
    image = cv2.imread("00.jpg")
    grayImage = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    equalHist = cv2.equalizeHist(grayImage, None)
    #自适应直方图均衡，参数是自适应的阈值
    adapthist = cv2.createCLAHE(clipLimit=40)
    imageAdaptHist = adapthist.apply(equalHist)
    cv2.imshow("adapthistimage",imageAdaptHist)
    cv2.imwrite("adapthist.jpg",imageAdaptHist)
    cv2.waitKey(0)

效果入下

上面照片比直方图均衡化后的图像，左下角好像没有那么白
通过对OpenCV提供的下面的一些参数，我们按照顺序统统试一下，看看有什么区别
cv2.THRESH_BINARY
cv2.THRESH_BINARY_INV
cv2.THRESH_TRUNC
cv2.THRESH_TOZERO
cv2.THRESH_TOZERO_INV
cv2.THRESH_OTSU
cv2.THRESH_TRIANGLE

#得到不同阈值
def Threshs():
    image = cv2.imread("00.jpg")
    grayImage = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    equalHist = cv2.equalizeHist(grayImage, None)
    # 自适应直方图均衡，参数是自适应的阈值
    adapthist = cv2.createCLAHE(clipLimit=40)
    imageAdaptHist = adapthist.apply(equalHist)
    thresh = 60
    max_val = 255
    ret, out1 = cv2.threshold(imageAdaptHist, thresh, max_val,cv2.THRESH_BINARY)
    ret, out2 = cv2.threshold(imageAdaptHist, thresh, max_val, cv2.THRESH_BINARY_INV)
    ret, out3 = cv2.threshold(imageAdaptHist, thresh, max_val, cv2.THRESH_TRUNC)
    ret, out4 = cv2.threshold(imageAdaptHist, thresh, max_val, cv2.THRESH_TOZERO)
    ret, out5 = cv2.threshold(imageAdaptHist, thresh, max_val,  cv2.THRESH_TOZERO_INV)
    ret, out6 = cv2.threshold(imageAdaptHist, thresh, max_val, cv2.THRESH_OTSU)
    ret, out7 = cv2.threshold(imageAdaptHist, thresh, max_val, cv2.THRESH_TRIANGLE)
    inal = np.concatenate((out1,out2,out3,out4), axis=0)
    inal1 = np.concatenate((out5,out6,out7), axis=0)
    cv2.imshow("test1",inal)
    cv2.imshow("test2",inal1)
    cv2.imwrite("test1.jpg",inal)
    cv2.imwrite("test2.jpg",inal1)
    cv2.waitKey(0)

嗯。。区别还是挺明显的，不过我最喜欢最后那个实验结果图，尽管有些不完整，最起码背景不那么复杂，比较好处理。
上面那些图像的高低阈值是我自己随便设置的，主观依赖性比较大，下面试试自适应阈值的方法

#自适应阈值处理
def AdaptThresh():
    grayimage = cv2.imread('00.jpg', 0)
    thresh1 = cv2.adaptiveThreshold(grayimage, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, 2)
    thresh2 = cv2.adaptiveThreshold(grayimage, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 31, 3)
    thresh3 = cv2.adaptiveThreshold(grayimage, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 13, 5)
    thresh4 = cv2.adaptiveThreshold(grayimage, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 31, 4)
    final = np.concatenate((thresh1, thresh2, thresh3, thresh4), axis=0)
    cv2.imshow("test1", final)
    cv2.imwrite("final.jpg", final)
    cv2.waitKey(0)

结果如下

好像不太行，不顾哦最后一张图像的效果已经很好了，著所以看着乱是因为背景是白色的原因，如果将图像减去255，效果立马就能感觉出来，这里我就不试了。

def result():
    gray_image = cv2.imread('00.jpg', 0)
    ret, thresh1 = cv2.threshold(gray_image, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
    cv2.imshow("test1", thresh1)
    cv2.imwrite("rest.jpg",thresh1)
    cv2.waitKey(0)

最后通过处理可以得到处理后的图像，如下所示

有干扰，尤其是左下角部分，但是整体上我们感兴趣的区域最起码可以看清了，而且效果还不错，应该不会影响对单个数字的分割，有兴趣的可以试试。