基于Python识别图像L形并提取其包围ROI的方法咨询

嘿，这个需求完全可以用OpenCV或者skimage轻松实现！我给你分享两种实用的思路，你可以根据自己图像的实际情况来选～

思路一：检测L形线条推导ROI边界 #

这种方法适合L形线条是ROI的两个邻边，另外两个边界需要通过线条延伸确定的场景，步骤如下：

用OpenCV实现的代码示例 #

import cv2
import numpy as np

# 1. 加载并预处理图像
img = cv2.imread('你的图像路径.png')
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# 边缘检测突出线条
edges = cv2.Canny(gray, 50, 150)

# 2. Hough直线检测找出所有线条
lines = cv2.HoughLinesP(edges, 1, np.pi/180, threshold=50, minLineLength=100, maxLineGap=10)

# 3. 分类水平和垂直线条
horizontal_lines = []
vertical_lines = []
for line in lines:
    x1, y1, x2, y2 = line[0]
    # 斜率接近0的是水平线条，斜率接近无穷的是垂直线条
    if abs(y2 - y1) < 10:
        horizontal_lines.append(line[0])
    elif abs(x2 - x1) < 10:
        vertical_lines.append(line[0])

# 4. 找到组成L形的相交线条
roi_lines = None
for h_line in horizontal_lines:
    h_x1, h_y, h_x2, _ = h_line
    for v_line in vertical_lines:
        v_x, v_y1, _, v_y2 = v_line
        # 判断两条线是否相交（L形的直角顶点）
        if (v_y1 <= h_y <= v_y2) and (h_x1 <= v_x <= h_x2):
            roi_lines = (h_line, v_line)
            break
    if roi_lines:
        break

# 5. 确定ROI边界并提取
if roi_lines:
    h_line, v_line = roi_lines
    h_x1, h_y, h_x2, _ = h_line
    v_x, v_y1, _, v_y2 = v_line

    # 这里根据L形的方向调整ROI范围，比如假设L是ROI的左上角边框，向右向下延伸
    # 你可以根据自己的图像实际情况修改x、y的上下限
    roi_x_start = v_x
    roi_x_end = h_x2  # 或者延伸到图像右侧/其他线条位置
    roi_y_start = h_y
    roi_y_end = v_y2  # 或者延伸到图像底部/其他线条位置

    # 提取ROI
    roi = img[roi_y_start:roi_y_end, roi_x_start:roi_x_end]
    cv2.imwrite('提取的ROI.png', roi)

用skimage实现的代码示例 #

from skimage import io, color, feature, transform
import numpy as np

# 1. 加载并预处理图像
img = io.imread('你的图像路径.png')
gray = color.rgb2gray(img)
edges = feature.canny(gray, sigma=1)

# 2. 概率Hough直线检测
lines = transform.probabilistic_hough_line(edges, threshold=50, line_length=100, line_gap=10)

# 3. 分类水平和垂直线条
horizontal_lines = []
vertical_lines = []
for line in lines:
    (x1, y1), (x2, y2) = line
    if abs(y2 - y1) < 10:
        horizontal_lines.append(line)
    elif abs(x2 - x1) < 10:
        vertical_lines.append(line)

# 4. 找到相交的L形线条
roi_lines = None
for h_line in horizontal_lines:
    (h_x1, h_y), (h_x2, _) = h_line
    for v_line in vertical_lines:
        (v_x, v_y1), (_, v_y2) = v_line
        if (v_y1 <= h_y <= v_y2) and (h_x1 <= v_x <= h_x2):
            roi_lines = (h_line, v_line)
            break
    if roi_lines:
        break

# 5. 提取ROI
if roi_lines:
    h_line, v_line = roi_lines
    (h_x1, h_y), (h_x2, _) = h_line
    (v_x, v_y1), (_, v_y2) = v_line

    # 调整ROI范围，根据你的图像实际情况修改
    roi_x_start = v_x
    roi_x_end = h_x2
    roi_y_start = h_y
    roi_y_end = v_y2

    roi = img[roi_y_start:roi_y_end, roi_x_start:roi_x_end]
    io.imsave('提取的ROI.png', roi)

思路二：直接检测矩形轮廓（适合表格类场景） #

如果你的图像是类似表格的结构，L形线条其实是矩形单元格的一部分边框，那直接检测矩形轮廓会更简单：

OpenCV代码示例 #

import cv2
import numpy as np

img = cv2.imread('你的图像路径.png')
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# 二值化+形态学操作，填充边框缝隙
_, binary = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY_INV)
kernel = np.ones((3,3), np.uint8)
binary = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel)

# 查找所有轮廓
contours, hierarchy = cv2.findContours(binary, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

# 筛选出符合条件的矩形轮廓
target_roi = None
for cnt in contours:
    area = cv2.contourArea(cnt)
    # 过滤太小的轮廓，阈值根据图像调整
    if area > 1000:
        # 近似轮廓为多边形
        approx = cv2.approxPolyDP(cnt, 0.02 * cv2.arcLength(cnt, True), True)
        # 四边形即为矩形候选
        if len(approx) == 4:
            x, y, w, h = cv2.boundingRect(cnt)
            # 可以根据位置/大小判断是否是目标ROI，比如打印坐标对比
            print(f"候选ROI坐标：x={x}, y={y}, w={w}, h={h}")
            # 假设第一个符合的是目标，或者你可以加判断条件
            target_roi = img[y:y+h, x:x+w]
            break

if target_roi is not None:
    cv2.imwrite('目标ROI.png', target_roi)

注意：所有代码里的参数（比如阈值、线条长度、面积阈值）都需要根据你的图像实际情况调整，多试几次就能找到最合适的参数啦～

内容的提问来源于stack exchange，提问作者Thalish Sajeed