图像处理那点事儿之二美轮美奂

- 日理万妓 2022-11-29 05:55 59阅读 0赞

接着上一篇图像处理那点事儿之凌波微步，我们今天来谈谈化妆术。你要了解颜色搭配，风格转换，提高你的审美，要不然女神会觉得被你喜欢是一件很丢人的事情。

## 2.图像美化 ##

### 1.直方图 ###

直方图可以很好的展示出图片中各种颜色分布的比重，对于我们分析图片有很大的帮助。实现起来也是很简单的，我们只需要把每种颜色值出现的比率计算出来，使用plt展示即可；有一点需要注意，颜色的计算需要分通道进行。  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2dhb2JpbmcxOTkz_size_16_color_FFFFFF_t_70_pic_center]

#原理，把0-255每个颜色所占像素点的数量用线段表示出来，不同的线段长度表示不用颜色的分布情况
    #使用opencv绘制彩色图像直方图
    #1.定义窗口名称 2.计算直方图信息 3.绘制直方图
    import cv2
    import  numpy as np
    #直方图计算逻辑
    def imageHist(image,type):
        color = (255,255,255)
        #1.定义显示窗口的名称 只有窗口名称不一致才可以同时展示多张图像
        windowName = 'Gray'
        if type == 31:
             color = (255,0,0)
             windowName = 'B Hist'
        elif type == 32:
             color = (0,255,0)
             windowName = 'G Hist'
        elif type == 33:
             color = (0,0,255)
             windowName = 'R Hist'
        #计算直方图信息
        # image channels mask histsize range 
        hist = cv2.calcHist([image],[0],None,[256],[0.0,255.0])
        #计算数组的最大最小值
        minV,maxV,minL,maxL = cv2.minMaxLoc(hist)
        print(maxV)
        #定义要展示的hist图像
        histImg = np.zeros([256,256,3],np.uint8)
        #3.绘制直方图信息
        for i in range(256):
            intenNormal =int(hist[i]*256/maxV)
            cv2.line(histImg,(i,256),(i,256-intenNormal),color)
        cv2.imshow(windowName,histImg)
        return histImg
    
    #遍历图像通道，调用方法
    img =cv2.imread('images/jinzhiyuan.jpg',1)
    #将图像按照通道进行分解RGB
    channels =  cv2.split(img);
    for i in range(0,3):
        imageHist(channels[i],i+31)
    cv2.waitKey(0)
    
    #灰色直方图源码
    #原理 求出各种颜色值所占比例大小，进行绘制
    import numpy as np
    import cv2
    import matplotlib.pyplot as plt
    img = cv2.imread('images/blue.jpg',0)
    imginfo = img.shape
    height = imginfo[0]
    width = imginfo[1]
    count = np.zeros(256,np.float)
    for i in range(height):
        for j in range(width):
            colorV = img[i,j]
            count[colorV] = count[colorV]+1
    for c in range(0,255):
        count[c] = count[c]/(height*width)
    x = np.linspace(0,255,256)
    y = count
    plt.bar(x,y,0.9,color='b')
    plt.show()
    
    #彩色直方图源码
    #原理 在灰度直方图的基础上，在rgb三个维度上进行统计 求出各种颜色值所占比例大小，进行绘制
    import numpy as np
    import cv2
    import matplotlib.pyplot as plt
    img = cv2.imread('images/tangwei.jpg',1)
    imginfo = img.shape
    height = imginfo[0]
    width = imginfo[1]
    countb = np.zeros(256,np.float)
    countg = np.zeros(256,np.float)
    countr = np.zeros(256,np.float)
    for i in range(height):
        for j in range(width):
            (b,g,r) = img[i,j]
            countb[b] = count[b]+1
            countg[g] = count[g]+1
            countr[r] = count[r]+1
    for c in range(0,255):
        countb[c] = countb[c]/(height*width)
        countg[c] = countg[c]/(height*width)
        countr[c] = countr[c]/(height*width)
    x = np.linspace(0,255,256)
    yb = countb
    plt.figure()
    plt.bar(x,yb,0.9,color='b')
    yg = countg
    plt.figure()
    plt.bar(x,yg,0.9,color='g')
    yr = countr
    plt.figure()
    plt.bar(x,yr,0.9,color='r')
    plt.show()

### 2.图片修补 ###

女神的照片上有一块缺失了怎么办？好着急啊。放心交给我吧，有我在你什么都不怕。图片修补是制作 一个只包含缺失块的图像蒙版，然后 调用cv2.inpaint，这个方法会对原图中蒙版的部分进行修补，修补的方式是采用半径为n的周围点的像素计算。

#1.对图像进行破坏
    import cv2
    import numpy as np
    img = cv2.imread('images/tongliya.jpg',1)
    for i in range(200,300):
        img[250,i] = (255,255,255)
        img[250+1,i] = (255,255,255)
        img[250-1,i] = (255,255,255)
    for j in range(200,300):
        img[j,250] = (255,255,255)
        img[j,250-1] = (255,255,255)
        img[j,250+1] = (255,255,255)
    cv2.imshow('img',img)
    cv2.imwrite('images/damage.jpg',img)
    cv2.waitKey(0)
    
    #图片修补
    #1.利用缺失的位置生成修补图片
    #2.调用cv2.inpaint进行修补
    import cv2
    import numpy as np
    img = cv2.imread('images/damage.jpg',1)
    img_info = img.shape
    height = img_info[0]
    width = img_info[1]
    mask_img = np.zeros([height,width,1],np.uint8)
    for i in range(200,300):
        mask_img[250,i] = 111
        mask_img[250+1,i] = 111
        mask_img[250-1,i] = 255
    for j in range(200,300):
        mask_img[j,250] = 1
        mask_img[j,250-1] = 255
        mask_img[j,250+1] = 255
    #INPAINT_TELEA
    #修补的原理是
    #mask_img 中不为0的区域作为img中待修补区域，3表示使用点附近半径为3的像素计算点的像素
    result = cv2.inpaint(img,mask_img,3,cv2.INPAINT_NS)
    cv2.imshow('img',img)
    cv2.imshow('mask_img',mask_img)
    cv2.imshow('result',result)
    cv2.waitKey(0)

\#\#\#3. 均衡化  
均衡化可以调整图像的对比度，对比度就是图片中最亮的白色到最暗的黑之间不同层级的测量。效果跟亮度的调整类似，但是和直接调整亮度不同的是，对比度调整是把亮度分成了很多层分别进行调整，调整后的效果是更深、更亮  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2dhb2JpbmcxOTkz_size_16_color_FFFFFF_t_70_pic_center 1]

#彩色图像均衡化 1.分通道 2.均衡化各通道 3合并
    import cv2
    import numpy as np
    img = cv2.imread('images/liqin.jpg',1)
    cv2.imshow('img',img)
    #分通道
    (b,g,r) = cv2.split(img)
    #均衡化各通道
    bH = cv2.equalizeHist(b)
    gH =  cv2.equalizeHist(g)
    rH = cv2.equalizeHist(r)
    #合并通道
    result = cv2.merge((bH,gH,rH))
    cv2.imshow('hist',result)
    cv2.waitKey(0)
    
    #彩色直方图均衡化源码
    #原理:就是在灰度均衡化基础上增加2个通道的处理
    #将不同层级的亮度分布，践行整体调整
    #步骤 1.计算累计概率 2.定义映射表 3.进行映射
    import numpy as np
    import cv2
    img = cv2.imread('images/tangwei.jpg',1)
    cv2.imshow('img',img)
    imginfo = img.shape
    height = imginfo[0]
    width = imginfo[1]
    countb = np.zeros(256,np.float)
    countg = np.zeros(256,np.float)
    countr = np.zeros(256,np.float)
    for i in range(height):
        for j in range(width):
            (b,g,r) = img[i,j]
            countb[b] = countb[b]+1
            countg[g] = countg[g]+1
            countr[r] = countr[r]+1
    for c in range(0,256):
        countb[c] = countb[c]/(height*width)
        countg[c] = countg[c]/(height*width)
        countr[c] = countr[c]/(height*width)
    #1.计算累计概率
    sumb = float(0)
    sumg = float(0)
    sumr = float(0)
    for i in range(0,256):
        sumb = sumb+countb[i]
        countb[i] = sumb
        sumg = sumg+countg[i]
        countg[i] = sumg
        sumr = sumr+countr[i]
        countr[i] = sumr
    #2. 定义映射表 映射关系为 x*255
    mapb = np.zeros(256,np.uint16)
    mapg = np.zeros(256,np.uint16)
    mapr = np.zeros(256,np.uint16)
    for i in range(0,256):
        mapb[i] = countb[i]*255
        mapg[i] = countg[i]*255
        mapr[i] = countr[i]*255
    for i in range(height):
        for j in range(width):
            (b,g,r) = img[i,j]
            b = mapb[b]
            g = mapg[g]
            r = mapr[r]
            img[i,j] = (b,g,r)  
    cv2.imshow('dest',img)
    cv2.waitKey(0)

### 4.亮度增强 ###

要让一个人变靚比登天还难，要让一张图变亮比睡觉还简单。把每一个像素值都变大就好了。  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2dhb2JpbmcxOTkz_size_16_color_FFFFFF_t_70_pic_center 2]

#亮度增强
    #原理：将每种颜色的值增大
    #跟对比度的区别是 对比度均衡化是按照比例，使黑的更黑，白的更白
    import cv2
    import numpy as np
    img = cv2.imread('images/tangwei.jpg',1)
    cv2.imshow('img',img)
    img_info = img.shape
    height = img_info[0]
    width = img_info[1]
    for i in range(0,height):
        for j in range(0,width):
            (b,g,r) = img[i,j]
            b_new = min(b*1.5+50,255)
            g_new = min(g*1.5+50,255)
            r_new = min(r*1.5+50,255)
            img[i,j] = (b_new,g_new,r_new)
    cv2.imshow('dest',img)
    cv2.waitKey(0)

### 5.磨皮美白 ###

磨皮美白会去掉原来图片中的噪声点，使照片看起来更干净、美好  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2dhb2JpbmcxOTkz_size_16_color_FFFFFF_t_70_pic_center 3]

#磨皮美白
    #使用双边滤波进行美白
    import cv2
    import numpy as np
    img = cv2.imread('images/liuhui2.jpg',1)
    cv2.imshow('img',img)
    #双边滤波
    dest = cv2.bilateralFilter(img,15,35,35)
    cv2.imshow('dest',dest)
    cv2.waitKey(0)

### 6.滤波器 ###

高斯滤波、均值滤波、中值滤波都是取出图像噪声点的方式。不过方式略有不同，均值滤波使用一个n\*n的滤波器点乘图像中的区域，使用点乘后的均值作为当前像素点的值。中值滤波也是点乘，使用点乘后的中位数作为当前像素点的值；高斯滤波的滤波器中每一个点的权重是不一样的，离目标点越近的点拥有越高的权值，然后进行加权平均

#高斯滤波
    import cv2
    import numpy as np
    img = cv2.imread('images/tangwei.jpg',1)
    cv2.imshow('img',img)
    #高斯滤波
    dest = cv2.GaussianBlur(img,(5,5),1.5)
    cv2.imwrite('images/gaussian.jpg',dest)
    cv2.imshow('dest',dest)
    cv2.waitKey(0)
    
    #均值滤波
    #原理 在n*n的区域内求均值，作为当前的值 相当于平均池化
    #步骤 相当于把每个点和它后面、下面的n个点加起来，求平均值作为这个点的值
    #对于每行每列最后n个值暂时没有进行处理
    import cv2
    import numpy as np
    img = cv2.imread('images/tangwei.jpg',1)
    cv2.imshow('img',img)
    img_info = img.shape
    height = img_info[0]
    width = img_info[1]
    #dest = np.zeros([height,width,3],np.uint8)
    for i in range(0,height-6):
        for j in range(0,width-6):
            sumb = int(0)
            sumg = int(0)
            sumr = int(0)
            for h in range(0,6):
                for w in range(0,6):
                    (b,g,r)  = img[i+h,j+w]
                    sumb+=int(b)
                    sumg+=int(g)
                    sumr+=int(r)
            img[i,j] = (sumb/36,sumg/36,sumr/36)
    cv2.imshow('dest',img)
    cv2.waitKey(0)
    
    #中值滤波
    #原理  在n*n的区域内求中间值，作为当前的值
    import cv2
    import numpy as np
    img = cv2.imread('images/liqin.jpg',1)
    cv2.imshow('img',img)
    img_info = img.shape
    height = img_info[0]
    width = img_info[1]
    #dest = np.zeros([height,width,3],np.uint8)
    collectb = np.zeros(9,np.uint8)
    collectg = np.zeros(9,np.uint8)
    collectr = np.zeros(9,np.uint8)
    for i in range(0,height-3):
        for j in range(0,width-3):
            k = 0;
            for h in range(0,3):
                for w in range(0,3):
                    (b,g,r)  = img[i+h,j+w]
                    collectb[k] = b
                    collectg[k] = g
                    collectr[k] = r
                    k+=1
            #冒泡排序
            for l in range(0,9):
                bl  =  collectb[l]
                gl  =  collectg[l]
                rl  =  collectr[l]
                for r in range(l+1,9):
                    if(bl<collectb[r]):
                        collectb[l] = collectb[r]
                        collectb[r] = bl
                    if(gl<collectg[r]):
                        collectg[l] = collectg[r]
                        collectg[r] = gl
                    if(rl<collectr[r]):
                        collectr[l] = collectr[r]
                        collectr[r] = rl
            b_middle = collectb[4]
            g_middle = collectg[4]
            r_middle = collectr[4]
            img[i,j] = (b_middle,g_middle,r_middle)       
    cv2.imshow('dest',img)
    cv2.waitKey(0)

图像美化章节且告一段落  
是我勇敢太久，决定为你一个人而活  
不能说出口那么折磨，勇敢了太久城市充满短暂的烟火  
无处躲，照亮了沉默  
明白是寂寞《勇敢-张惠妹》  
如果喜欢，请继续观看下一篇《图像处理那点事儿之三特异功能》

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2dhb2JpbmcxOTkz_size_16_color_FFFFFF_t_70_pic_center]: /images/20221124/e4ab79e3462742c9bdbb9f37933a3017.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2dhb2JpbmcxOTkz_size_16_color_FFFFFF_t_70_pic_center 1]: /images/20221124/5b0ac7e2136641a6a7db92a05309868b.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2dhb2JpbmcxOTkz_size_16_color_FFFFFF_t_70_pic_center 2]: /images/20221124/90736732218847eea64621657d6dd147.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2dhb2JpbmcxOTkz_size_16_color_FFFFFF_t_70_pic_center 3]: /images/20221124/b66c825f15094816b821cdd9516aa401.png