python 图像分割方法总结

图像分割是一种常用的图像处理方法，可分为传统方法和深度学习的方法。深度学习的方法比如：mask rcnn这类实例分割模型，效果比传统的图像分割方法要好的多，所以目前图像分割领域都是用深度学习来做的。但是深度学习也有它的缺点，模型大、推理速度慢、可解释性差、训练数据要求高等。本文在这里仅讨论传统的图像分割算法，可供学习和使用。
1、阈值分割
最简单的图像分割算法，只直接按照像素值进行分割，虽然简单，但是在一些像素差别较大的场景中表现不错，是一种简单而且稳定的算法。

def thresholdSegment(filename):
    gray = cv2.imread(filename, cv2.IMREAD_GRAYSCALE)
    ret1, th1 = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY)
    th2 = cv2.adaptiveThreshold(
        gray, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, 2)
    th3 = cv2.adaptiveThreshold(
        gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
    ret2, th4 = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
    images = [th1, th2, th4, th3]
    imgaesTitle = ['THRESH_BINARY', 'THRESH_MEAN',
                   'THRESH_OTSU', 'THRESH_GAUSSIAN']
    plt.figure()
    for i in range(4):
        plt.subplot(2, 2, i+1)
        plt.imshow(images[i], 'gray')
        plt.title(imgaesTitle[i])
        cv2.imwrite(imgaesTitle[i]+'.jpg', images[i])
    plt.show()
    cv2.waitKey(0)
    return images

2、边界分割（边缘检测）

def edgeSegmentation(filename):
    # 读取图片
    img = cv2.imread(filename)
    # 灰度化
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    # 高斯模糊处理:去噪(效果最好)
    blur = cv2.GaussianBlur(gray, (9, 9), 0)
    # Sobel计算XY方向梯度
    gradX = cv2.Sobel(gray, ddepth=cv2.CV_32F, dx=1, dy=0)
    gradY = cv2.Sobel(gray, ddepth=cv2.CV_32F, dx=0, dy=1)
    # 计算梯度差
    gradient = cv2.subtract(gradX, gradY)
    # 绝对值
    gradient = cv2.convertScaleAbs(gradient)
    # 高斯模糊处理:去噪(效果最好)
    blured = cv2.GaussianBlur(gradient, (9, 9), 0)
    # 二值化
    _, dst = cv2.threshold(blured, 90, 255, cv2.THRESH_BINARY)
    # 滑动窗口
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (107, 76))
    # 形态学处理:形态闭处理(腐蚀)
    closed = cv2.morphologyEx(dst, cv2.MORPH_CLOSE, kernel)
    # 腐蚀与膨胀迭代
    closed = cv2.erode(closed, None, iterations=4)
    closed = cv2.dilate(closed, None, iterations=4)

    # 获取轮廓
    _, cnts, _ = cv2.findContours(
        closed.copy(), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
    c = sorted(cnts, key=cv2.contourArea, reverse=True)[0]
    rect = cv2.minAreaRect(c)
    box = np.int0(cv2.boxPoints(rect))
    draw_img = cv2.drawContours(img.copy(), [box], -1, (0, 0, 255), 3)
    #cv2.imshow("Box", draw_img)
    #cv2.imwrite('./test/monkey.png', draw_img)
    images = [blured, dst, closed, draw_img]
    imgaesTitle = ['blured', 'dst', 'closed', 'draw_img']
    plt.figure()
    for i in range(4):
        plt.subplot(2, 2, i+1)
        plt.imshow(images[i], 'gray')
        plt.title(imgaesTitle[i])
        #cv2.imwrite(imgaesTitle[i]+'.jpg', images[i])
    plt.show()
    cv2.waitKey(0)

3、区域分割（区域生成）

def regionSegmentation(filename):
    # 读取图片
    img = cv2.imread(filename)
    # 图片宽度
    img_x = img.shape[1]
    # 图片高度
    img_y = img.shape[0]
    # 分割的矩形区域
    rect = (0, 0, img_x-1, img_y-1)
    # 背景模式,必须为1行,13x5列
    bgModel = np.zeros((1, 65), np.float64)
    # 前景模式,必须为1行,13x5列
    fgModel = np.zeros((1, 65), np.float64)
    # 图像掩模,取值有0,1,2,3
    mask = np.zeros(img.shape[:2], np.uint8)
    # grabCut处理,GC_INIT_WITH_RECT模式
    cv2.grabCut(img, mask, rect, bgModel, fgModel, 4, cv2.GC_INIT_WITH_RECT)
    # grabCut处理,GC_INIT_WITH_MASK模式
    #cv2.grabCut(img, mask, rect, bgModel, fgModel, 4, cv2.GC_INIT_WITH_MASK)
    # 将背景0,2设成0,其余设成1
    mask2 = np.where((mask == 2) | (mask == 0), 0, 1).astype('uint8')
    # 重新计算图像着色,对应元素相乘
    img = img*mask2[:, :, np.newaxis]
    cv2.imshow("Result", img)
    cv2.waitKey(0)

4、SVM分割（支持向量机）

def svmSegment(pic):
    img = Image.open(pic)
    img.show()  # 显示原始图像
    img_arr = np.asarray(img, np.float64)
  #选取图像上的关键点RGB值(10个)
    lake_RGB = np.array(
    [[147, 168, 125], [151, 173, 124], [143, 159, 112], [150, 168, 126], [146, 165, 120],
     [145, 161, 116], [150, 171, 130], [146, 112, 137], [149, 169, 120], [144, 160, 111]])
# 选取待分割目标上的关键点RGB值(10个)
    duck_RGB = np.array(
    [[81, 76, 82], [212, 202, 193], [177, 159, 157], [129, 112, 105], [167, 147, 136],
     [237, 207, 145], [226, 207, 192], [95, 81, 68], [198, 216, 218], [197, 180, 128]] )
    RGB_arr = np.concatenate((lake_RGB, duck_RGB), axis=0)  # 按列拼接
    # lake 用 0标记，duck用1标记
    label = np.append(np.zeros(lake_RGB.shape[0]), np.ones(duck_RGB.shape[0]))
    # 原本 img_arr 形状为(m,n,k),现在转化为(m*n,k)
    img_reshape = img_arr.reshape(
    [img_arr.shape[0]*img_arr.shape[1], img_arr.shape[2]])
    svc = SVC(kernel='poly', degree=3)  # 使用多项式核，次数为3
    svc.fit(RGB_arr, label)  # SVM 训练样本
    predict = svc.predict(img_reshape)  # 预测测试点
    lake_bool = predict == 0.  
    lake_bool = lake_bool[:, np.newaxis]  # 增加一列(一维变二维)
    lake_bool_3col = np.concatenate(
    (lake_bool, lake_bool, lake_bool), axis=1)  # 变为三列
    lake_bool_3d = lake_bool_3col.reshape(
    (img_arr.shape[0], img_arr.shape[1], img_arr.shape[2]))  # 变回三维数组(逻辑数组)
    img_arr[lake_bool_3d] = 255.  
    img_split = Image.fromarray(img_arr.astype('uint8'))  # 数组转image
    img_split.show()  # 显示分割之后的图像
    img_split.save('split_duck.jpg')  # 保存

5、分水岭分割

def watershedSegment(filename):
    img = cv2.imread(filename)
    gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
    ret, thresh = cv2.threshold(gray,0,255,cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
    # noise removal
    kernel = np.ones((3,3),np.uint8)
    opening = cv2.morphologyEx(thresh,cv2.MORPH_OPEN,kernel, iterations = 2)

    # sure background area
    sure_bg = cv2.dilate(opening,kernel,iterations=3)

    # Finding sure foreground area
    dist_transform = cv2.distanceTransform(opening,cv2.DIST_L2,5)
    ret, sure_fg = cv2.threshold(dist_transform,0.7*dist_transform.max(),255,0)

    # Finding unknown region
    sure_fg = np.uint8(sure_fg)
    unknown = cv2.subtract(sure_bg,sure_fg) 
    # Marker labelling
    ret, markers = cv2.connectedComponents(sure_fg)

    # Add one to all labels so that sure background is not 0, but 1
    markers = markers+1

    # Now, mark the region of unknown with zero
    markers[unknown==255]=0 
    markers = cv2.watershed(img,markers)
    img[markers == -1] = [255,0,0]

6、Kmeans分割

def kmeansSegment(filename,k):
    f = open(filename,'rb') #二进制打开
    data = []
    img = Image.open(f) #以列表形式返回图片像素值
    m,n = img.size #图片大小
    for i in range(m):
        for j in range(n):  #将每个像素点RGB颜色处理到0-1范围内并存放data
            x,y,z = img.getpixel((i,j))
            data.append([x/256.0,y/256.0,z/256.0])
    f.close()
    img_data=np.mat(data)
    row=m
    col=n
    label = KMeans(n_clusters=k).fit_predict(img_data)  #聚类中心的个数为3
    label = label.reshape([row,col])    #聚类获得每个像素所属的类别
    pic_new = Image.new("L",(row,col))  #创建一张新的灰度图保存聚类后的结果
    for i in range(row):    #根据所属类别向图片中添加灰度值
        for j in range(col):
            pic_new.putpixel((i,j),int(256/(label[i][j]+1)))
    pic_new.save('keans_'+str(k)+'.jpg')
    plt.imshow(pic_new)
    plt.show()