如何检测非直线？解决cv2.HoughLinesP检测不全及图像分割问题

1. 优化HoughLinesP参数提升检测效果 #

你当前的参数设置可能导致线条检测不全或断裂，调整以下参数可改善结果：

• threshold：适度提高阈值过滤噪声（示例调整为50）
• minLineLength：设置最小线段长度，过滤短噪线段（示例设为100）
• maxLineGap：允许同一直线上线段的最大间隙，让相近线段归为一组（示例设为20）

调整后的检测代码：

import cv2
import numpy as np

img = cv2.imread("testImages.png")
img_copy = img.copy()
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# 先做高斯模糊减少噪声，提升Canny边缘检测效果
img_blur = cv2.GaussianBlur(img_gray, (3,3), 0)
edges = cv2.Canny(img_blur, 100, 200, apertureSize=3)
# 调整HoughLinesP参数
lines = cv2.HoughLinesP(edges, cv2.HOUGH_PROBABILISTIC, np.pi/180, threshold=50, minLineLength=100, maxLineGap=20)

2. 合并同一直线的断裂线段 #

通过斜率一致性和端点距离判断线段是否属于同一直线，合并为完整线段：

def merge_lines(lines, angle_threshold=0.1, distance_threshold=20):
    merged_lines = []
    if lines is None:
        return merged_lines
    
    # 预处理线段，计算斜率和截距
    line_info = []
    for line in lines:
        x1, y1, x2, y2 = line[0]
        # 处理垂直线斜率无穷大的情况
        if x2 - x1 == 0:
            slope = np.inf
            intercept = x1
        else:
            slope = (y2 - y1) / (x2 - x1)
            intercept = y1 - slope * x1
        line_info.append((x1, y1, x2, y2, slope, intercept))
    
    # 遍历合并相似线段
    used = [False]*len(line_info)
    for i in range(len(line_info)):
        if used[i]:
            continue
        current = line_info[i]
        x1, y1, x2, y2, s1, int1 = current
        group = [current]
        used[i] = True
        
        for j in range(i+1, len(line_info)):
            if used[j]:
                continue
            x3, y3, x4, y4, s2, int2 = line_info[j]
            # 检查斜率角度差
            angle_diff = abs(np.arctan(s1) - np.arctan(s2))
            if angle_diff > np.pi - angle_threshold:
                angle_diff = 2*np.pi - angle_diff
            if angle_diff > angle_threshold:
                continue
            # 检查截距或端点距离
            if abs(int1 - int2) > distance_threshold:
                dists = [np.hypot(x1-x3, y1-y3), np.hypot(x1-x4, y1-y4),
                         np.hypot(x2-x3, y2-y3), np.hypot(x2-x4, y2-y4)]
                if min(dists) > distance_threshold:
                    continue
            group.append(line_info[j])
            used[j] = True
        
        # 合并组内线段，取极值端点
        all_x = []
        all_y = []
        for l in group:
            all_x.extend([l[0], l[2]])
            all_y.extend([l[1], l[3]])
        min_x, max_x = min(all_x), max(all_x)
        min_y, max_y = min(all_y), max(all_y)
        
        # 根据斜率判断直线类型，生成合并后的线段
        if abs(s1) < 0.1:  # 水平线
            merged_lines.append((min_x, min_y, max_x, min_y))
        elif abs(s1) > 10:  # 垂直线
            merged_lines.append((min_x, min_y, min_x, max_y))
        else:
            merged_lines.append((min_x, min_y, max_x, max_y))
    return merged_lines

# 调用合并函数
merged_lines = merge_lines(lines)

3. 基于合并后的线条分割图像 #

假设图像由水平/垂直线分割为网格，提取线条坐标后裁剪：

def split_image(img, merged_lines):
    horizontal_ys = []
    vertical_xs = []
    # 分离水平线和垂直线坐标
    for line in merged_lines:
        x1, y1, x2, y2 = line
        if abs(y1 - y2) < 5:  # 水平线
            horizontal_ys.append(y1)
        elif abs(x1 - x2) < 5:  # 垂直线
            vertical_xs.append(x1)
    
    # 排序去重，添加图像边界
    horizontal_ys = sorted(list(set([round(y) for y in horizontal_ys])))
    vertical_xs = sorted(list(set([round(x) for x in vertical_xs])))
    horizontal_ys.insert(0, 0)
    horizontal_ys.append(img.shape[0])
    vertical_xs.insert(0, 0)
    vertical_xs.append(img.shape[1])
    
    # 裁剪每个区域，跳过过小的无效区域
    split_imgs = []
    for i in range(len(horizontal_ys)-1):
        y_start, y_end = horizontal_ys[i], horizontal_ys[i+1]
        for j in range(len(vertical_xs)-1):
            x_start, x_end = vertical_xs[j], vertical_xs[j+1]
            if (y_end - y_start) < 10 or (x_end - x_start) < 10:
                continue
            crop_img = img[y_start:y_end, x_start:x_end]
            split_imgs.append(crop_img)
    return split_imgs

# 执行分割并保存结果
split_images = split_image(img, merged_lines)
for idx, crop in enumerate(split_images):
    cv2.imwrite(f"crop_{idx}.png", crop)

额外提示 #

• 若存在倾斜线条，需调整合并逻辑，改用直线参数方程判断
• 可先对图像做二值化处理，强化线条对比度，提升检测准确性
• 合并线段的阈值（角度、距离）需根据实际图像微调，直到得到完整线条

内容的提问来源于stack exchange，提问作者Albert