OpenCV-Python实战|图像分割全解析:分水岭算法+GrabCut交互式前景提取(附完整代码)

python# 安装/升级OpenCV-Python（核心依赖）pip install opencv-python -U# 安装辅助库（图像显示、数据处理、交互式操作）pip install numpy matplotlib

pythonimport cv2import numpy as npimport matplotlib.pyplot as plt# 1. 读取两类测试图（灰度图/彩色图，根据算法需求选择）# 测试图1：粘连目标图（用于分水岭算法，灰度图）img_watershed = cv2.imread("watershed_test.jpg", 0) # 替换为自己的图像路径# 测试图2：前景背景混合图（用于GrabCut算法，彩色图）img_grabcut = cv2.imread("grabcut_test.jpg") # 替换为自己的图像路径img_grabcut_gray = cv2.cvtColor(img_grabcut, cv2.COLOR_BGR2GRAY)# 2. 图像预处理：高斯平滑（降噪），衔接前文技巧img_watershed_blur = cv2.GaussianBlur(img_watershed, (3, 3), sigmaX=1)img_grabcut_blur = cv2.GaussianBlur(img_grabcut, (3, 3), sigmaX=1)# 显示测试图plt.figure(figsize=(10, 5))# 粘连目标图plt.subplot(1, 2, 1)plt.imshow(img_watershed_blur, cmap="gray")plt.title("粘连目标图（分水岭测试图）")plt.axis("off")# 前景背景混合图plt.subplot(1, 2, 2)plt.imshow(cv2.cvtColor(img_grabcut_blur, cv2.COLOR_BGR2RGB))plt.title("前景背景混合图（GrabCut测试图）")plt.axis("off")plt.show()

pythonimport cv2import numpy as npimport matplotlib.pyplot as plt# 1. 读取图像并预处理（平滑降噪）img = cv2.imread("watershed_test.jpg", 0)img_blur = cv2.GaussianBlur(img, (3, 3), sigmaX=1)# 2. 二值化处理（突出目标，阈值可调整）# 反二值化：目标为白色（255），背景为黑色（0）ret, thresh = cv2.threshold(img_blur, 127, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)# 3. 形态学操作：去除噪声，填补目标内部孔洞（衔接前文形态学知识点）kernel = np.ones((3, 3), np.uint8)# 开运算：先腐蚀再膨胀，去除小噪声opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2)# 膨胀操作：填补目标内部孔洞，得到背景区域sure_bg = cv2.dilate(opening, kernel, iterations=3)# 4. 距离变换：计算前景内部点到背景的距离，用于寻找种子点dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 5)# 阈值处理，得到确定的前景区域（种子点区域）ret, sure_fg = cv2.threshold(dist_transform, 0.7*dist_transform.max(), 255, 0)sure_fg = np.uint8(sure_fg)# 5. 计算不确定区域（前景与背景之间的区域）unknown = cv2.subtract(sure_bg, sure_fg)# 6. 标记种子点（为每个前景目标分配唯一标签）ret, markers = cv2.connectedComponents(sure_fg)# 所有标签+1，让背景标签从0变为1（避免与分水岭标记冲突）markers = markers + 1# 将不确定区域标记为0（让算法自动分割该区域）markers[unknown == 255] = 0# 7. 执行分水岭分割（需将灰度图转为彩色图，用于绘制分割线）img_color = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)markers = cv2.watershed(img_color, markers)# 绘制分水岭分割线（标记为红色，厚度1）img_color[markers == -1] = [0, 0, 255]# 显示效果（预处理→二值化→分割结果）plt.figure(figsize=(12, 4))# 预处理后的图像plt.subplot(1, 3, 1)plt.imshow(img_blur, cmap="gray")plt.title("预处理后图像")plt.axis("off")# 二值化图像plt.subplot(1, 3, 2)plt.imshow(thresh, cmap="gray")plt.title("二值化图像")plt.axis("off")# 分水岭分割结果plt.subplot(1, 3, 3)plt.imshow(cv2.cvtColor(img_color, cv2.COLOR_BGR2RGB))plt.title("分水岭算法分割结果")plt.axis("off")plt.tight_layout()plt.show()# 统计分割后的目标数量（标签从2开始，因为背景为1，不确定区域为0）target_count = len(np.unique(markers)) - 2print(f"共分割出{target_count}个目标")

pythoncv2.grabCut(img, mask, rect, bgdModel, fgdModel, iterCount, mode)

pythonimport cv2import numpy as npimport matplotlib.pyplot as plt# 1. 读取彩色图像并预处理（平滑降噪）img = cv2.imread("grabcut_test.jpg")img_blur = cv2.GaussianBlur(img, (3, 3), sigmaX=1)# 复制原始图像，用于绘制矩形框img_copy = img_blur.copy()# 2. 手动框选前景区域（x, y, w, h），需根据自己的图像调整坐标# 示例：x=50, y=50, w=300, h=400（可通过图像查看工具获取坐标）rect = (50, 50, 300, 400)# 3. 初始化掩码、背景模型、前景模型mask = np.zeros(img.shape[:2], np.uint8) # 掩码初始全0bgdModel = np.zeros((1, 65), np.float64) # 背景模型（固定格式）fgdModel = np.zeros((1, 65), np.float64) # 前景模型（固定格式）# 4. 执行GrabCut算法cv2.grabCut(img_blur,mask,rect,bgdModel,fgdModel,iterCount=5, # 迭代5次，兼顾精度和速度mode=cv2.GC_INIT_WITH_RECT # 基于矩形框初始化)# 5. 处理掩码，提取前景区域# 掩码值说明：0=背景，1=前景，2=不确定背景，3=不确定前景# 将不确定背景（2）和确定背景（0）设为0，确定前景（1）和不确定前景（3）设为255mask2 = np.where((mask == 2) | (mask == 0), 0, 255)# 将掩码转为3通道，与原始图像叠加，得到前景提取结果foreground = cv2.bitwise_and(img, img, mask=mask2.astype(np.uint8))# 6. 绘制手动框选的矩形框（用于展示）cv2.rectangle(img_copy, (rect[0], rect[1]), (rect[0]+rect[2], rect[1]+rect[3]), (0, 255, 0), 2)cv2.putText(img_copy, "Foreground Area", (rect[0], rect[1]-10), cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0, 255, 0), 2)# 显示效果（原始图+框选→掩码图→前景提取结果）plt.figure(figsize=(12, 4))# 原始图+手动框选plt.subplot(1, 3, 1)plt.imshow(cv2.cvtColor(img_copy, cv2.COLOR_BGR2RGB))plt.title("原始图+前景框选")plt.axis("off")# 掩码图plt.subplot(1, 3, 2)plt.imshow(mask2, cmap="gray")plt.title("掩码图（白色=前景）")plt.axis("off")# 前景提取结果plt.subplot(1, 3, 3)plt.imshow(cv2.cvtColor(foreground, cv2.COLOR_BGR2RGB))plt.title("GrabCut前景提取结果")plt.axis("off")plt.tight_layout()plt.show()# 保存前景提取结果（可选）cv2.imwrite("grabcut_foreground.jpg", foreground)

pythonimport cv2import numpy as npimport matplotlib.pyplot as plt# 第一步：用分水岭算法分割粘连目标img_watershed = cv2.imread("watershed_test.jpg", 0)img_watershed_blur = cv2.GaussianBlur(img_watershed, (3, 3), sigmaX=1)# 二值化+形态学操作ret, thresh = cv2.threshold(img_watershed_blur, 127, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)kernel = np.ones((3, 3), np.uint8)opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2)sure_bg = cv2.dilate(opening, kernel, iterations=3)# 距离变换+种子点标记dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 5)ret, sure_fg = cv2.threshold(dist_transform, 0.7*dist_transform.max(), 255, 0)sure_fg = np.uint8(sure_fg)unknown = cv2.subtract(sure_bg, sure_fg)ret, markers = cv2.connectedComponents(sure_fg)markers = markers + 1markers[unknown == 255] = 0# 分水岭分割img_watershed_color = cv2.cvtColor(img_watershed, cv2.COLOR_GRAY2BGR)markers = cv2.watershed(img_watershed_color, markers)img_watershed_color[markers == -1] = [0, 0, 255]# 第二步：用GrabCut算法提取单个分割目标（以第一个目标为例）# 找到第一个目标的坐标范围（根据标记值筛选）target_label = 2 # 第一个目标的标记值（从2开始）target_mask = np.where(markers == target_label, 255, 0)target_mask = np.uint8(target_mask)# 找到目标的边界框（用于GrabCut框选）contours, _ = cv2.findContours(target_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)x, y, w, h = cv2.boundingRect(contours[0])rect = (x, y, w, h)# 读取彩色版粘连目标图（用于GrabCut提取）img_color = cv2.imread("watershed_test.jpg")img_color_blur = cv2.GaussianBlur(img_color, (3, 3), sigmaX=1)# 执行GrabCut提取单个目标mask_grabcut = np.zeros(img_color.shape[:2], np.uint8)bgdModel = np.zeros((1, 65), np.float64)fgdModel = np.zeros((1, 65), np.float64)cv2.grabCut(img_color_blur, mask_grabcut, rect, bgdModel, fgdModel, iterCount=5, mode=cv2.GC_INIT_WITH_RECT)# 处理掩码，得到单个目标前景mask_grabcut2 = np.where((mask_grabcut == 2) | (mask_grabcut == 0), 0, 255)single_foreground = cv2.bitwise_and(img_color, img_color, mask=mask_grabcut2.astype(np.uint8))# 显示综合应用效果plt.figure(figsize=(15, 5))# 原始粘连目标图plt.subplot(1, 3, 1)plt.imshow(cv2.cvtColor(img_color, cv2.COLOR_BGR2RGB))plt.title("原始粘连目标图")plt.axis("off")# 分水岭分割结果plt.subplot(1, 3, 2)plt.imshow(cv2.cvtColor(img_watershed_color, cv2.COLOR_BGR2RGB))plt.title("分水岭分割结果（粘连目标分离）")plt.axis("off")# GrabCut提取单个目标plt.subplot(1, 3, 3)plt.imshow(cv2.cvtColor(single_foreground, cv2.COLOR_BGR2RGB))plt.title("GrabCut提取单个目标前景")plt.axis("off")plt.tight_layout()plt.show()# 输出结果统计target_count = len(np.unique(markers)) - 2print(f"共分割出{target_count}个粘连目标，成功提取1个目标前景")

pythonimport cv2import numpy as npimport matplotlib.pyplot as plt# 1. 环境验证（可选）print("OpenCV版本：", cv2.__version__)# 2. 读取图像（替换为自己的图像路径）# 分水岭测试图（粘连目标，灰度图）img_watershed = cv2.imread("watershed_test.jpg", 0)# GrabCut测试图（前景背景混合，彩色图）img_grabcut = cv2.imread("grabcut_test.jpg")if img_watershed is None or img_grabcut is None:print("错误：未找到图像，请检查图像路径！")else:# 3. 分水岭算法：分割粘连目标# 预处理img_watershed_blur = cv2.GaussianBlur(img_watershed, (3, 3), sigmaX=1)# 二值化+形态学操作ret, thresh = cv2.threshold(img_watershed_blur, 127, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)kernel = np.ones((3, 3), np.uint8)opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2)sure_bg = cv2.dilate(opening, kernel, iterations=3)# 距离变换+种子点标记dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 5)ret, sure_fg = cv2.threshold(dist_transform, 0.7*dist_transform.max(), 255, 0)sure_fg = np.uint8(sure_fg)unknown = cv2.subtract(sure_bg, sure_fg)ret, markers = cv2.connectedComponents(sure_fg)markers = markers + 1markers[unknown == 255] = 0# 执行分水岭分割img_watershed_color = cv2.cvtColor(img_watershed, cv2.COLOR_GRAY2BGR)markers = cv2.watershed(img_watershed_color, markers)img_watershed_color[markers == -1] = [0, 0, 255]# 统计分割目标数量target_count = len(np.unique(markers)) - 2# 4. GrabCut算法：交互式前景提取# 预处理img_grabcut_blur = cv2.GaussianBlur(img_grabcut, (3, 3), sigmaX=1)img_grabcut_copy = img_grabcut_blur.copy()# 手动框选前景区域（根据自己的图像调整坐标）rect = (50, 50, 300, 400)# 初始化掩码和模型mask_grabcut = np.zeros(img_grabcut.shape[:2], np.uint8)bgdModel = np.zeros((1, 65), np.float64)fgdModel = np.zeros((1, 65), np.float64)# 执行GrabCutcv2.grabCut(img_grabcut_blur, mask_grabcut, rect, bgdModel, fgdModel, iterCount=5, mode=cv2.GC_INIT_WITH_RECT)# 处理掩码，提取前景mask_grabcut2 = np.where((mask_grabcut == 2) | (mask_grabcut == 0), 0, 255)foreground = cv2.bitwise_and(img_grabcut, img_grabcut, mask=mask_grabcut2.astype(np.uint8))# 绘制矩形框cv2.rectangle(img_grabcut_copy, (rect[0], rect[1]), (rect[0]+rect[2], rect[1]+rect[3]), (0, 255, 0), 2)cv2.putText(img_grabcut_copy, "Foreground Area", (rect[0], rect[1]-10), cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0, 255, 0), 2)# 5. 综合应用：分水岭分割+GrabCut提取单个目标# 读取彩色版粘连目标图img_color = cv2.imread("watershed_test.jpg")img_color_blur = cv2.GaussianBlur(img_color, (3, 3), sigmaX=1)# 找到第一个目标的边界框target_label = 2target_mask = np.where(markers == target_label, 255, 0)target_mask = np.uint8(target_mask)contours, _ = cv2.findContours(target_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)x, y, w, h = cv2.boundingRect(contours[0])rect_single = (x, y, w, h)# 执行GrabCut提取单个目标mask_single = np.zeros(img_color.shape[:2], np.uint8)bgdModel_single = np.zeros((1, 65), np.float64)fgdModel_single = np.zeros((1, 65), np.float64)cv2.grabCut(img_color_blur, mask_single, rect_single, bgdModel_single, fgdModel_single, iterCount=5, mode=cv2.GC_INIT_WITH_RECT)mask_single2 = np.where((mask_single == 2) | (mask_single == 0), 0, 255)single_foreground = cv2.bitwise_and(img_color, img_color, mask=mask_single2.astype(np.uint8))# 6. 统一显示所有效果plt.figure(figsize=(15, 10))# 第一行：分水岭相关plt.subplot(2, 3, 1)plt.imshow(img_watershed_blur, cmap="gray")plt.title("分水岭预处理图")plt.axis("off")plt.subplot(2, 3, 2)plt.imshow(thresh, cmap="gray")plt.title("分水岭二值化图")plt.axis("off")plt.subplot(2, 3, 3)plt.imshow(cv2.cvtColor(img_watershed_color, cv2.COLOR_BGR2RGB))plt.title(f"分水岭分割（{target_count}个目标）")plt.axis("off")# 第二行：GrabCut相关+综合应用plt.subplot(2, 3, 4)plt.imshow(cv2.cvtColor(img_grabcut_copy, cv2.COLOR_BGR2RGB))plt.title("GrabCut框选前景")plt.axis("off")plt.subplot(2, 3, 5)plt.imshow(cv2.cvtColor(foreground, cv2.COLOR_BGR2RGB))plt.title("GrabCut前景提取")plt.axis("off")plt.subplot(2, 3, 6)plt.imshow(cv2.cvtColor(single_foreground, cv2.COLOR_BGR2RGB))plt.title("综合应用：提取单个分割目标")plt.axis("off")plt.tight_layout()plt.show()# 7. 保存结果（可选）cv2.imwrite("watershed_result.jpg", img_watershed_color)cv2.imwrite("grabcut_result.jpg", foreground)cv2.imwrite("single_target_result.jpg", single_foreground)