Python提取拼接图中子图：现有OpenCV代码优化及替代方案咨询

现有代码的针对性优化 #

你的代码漏检主要是因为边缘不连续、阈值固定、轮廓筛选条件太严格，针对这些问题可以逐个调整：

1. 补全断边，优化边缘检测流程 #

Canny边缘检测容易出现断边，加上形态学运算和模糊去噪可以解决：

def detect_objects(frame):
    gray_image = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    # 高斯模糊去噪，减少Canny的误检
    blur = cv2.GaussianBlur(gray_image, (5, 5), 0)
    # 调整Canny阈值（降低高阈值，让更多边缘被保留）
    edges = cv2.Canny(blur, 30, 100)
    # 闭运算补全边缘缺口，用3x3矩形核连接断开的边缘
    kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
    closed_edges = cv2.morphologyEx(edges, cv2.MORPH_CLOSE, kernel)

    # 找轮廓时改用RETR_LIST，避免漏检内部轮廓
    contours, _ = cv2.findContours(
        closed_edges.astype(np.uint8), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE
    )

    # 动态计算最小面积阈值，适配不同尺寸的图片
    img_area = frame.shape[0] * frame.shape[1]
    min_area = img_area * 0.0001  # 可根据子图大小调整比例

    for i, contour in enumerate(contours):
        area = cv2.contourArea(contour)
        if area < min_area:
            continue
        # 放宽多边形逼近的epsilon参数，避免把接近矩形的轮廓排除
        epsilon = 0.03 * cv2.arcLength(contour, True)
        approx = cv2.approxPolyDP(contour, epsilon, True)
        # 允许近似四边形（4或5个顶点，适配边缘略有变形的情况）
        if len(approx) in [4, 5]:
            # 额外判断外接矩形的宽高比，过滤非子图的矩形
            x, y, w, h = cv2.boundingRect(approx)
            aspect_ratio = float(w) / h
            # 假设子图接近正方形，可根据实际情况调整比例范围
            if 0.7 < aspect_ratio < 1.3:
                cv2.drawContours(frame, [approx], -1, (0, 255, 0), 3)
                print(f"Area of contour {i+1}: {area}")

    return frame

2. 替换边缘检测为自适应阈值分割 #

如果子图和背景的灰度差异明显，直接用自适应阈值分割比Canny更稳定：

def detect_objects(frame):
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    # 自适应阈值分割，自动区分子图和背景
    thresh = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
                                   cv2.THRESH_BINARY_INV, 11, 2)
    # 开运算去掉小噪点，避免误检小轮廓
    kernel = np.ones((3, 3), np.uint8)
    thresh = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel)

    contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    img_area = frame.shape[0] * frame.shape[1]
    min_area = img_area * 0.0001

    for i, contour in enumerate(contours):
        area = cv2.contourArea(contour)
        if area < min_area:
            continue
        epsilon = 0.03 * cv2.arcLength(contour, True)
        approx = cv2.approxPolyDP(contour, epsilon, True)
        if len(approx) in [4,5]:
            cv2.drawContours(frame, [approx], -1, (0,255,0), 3)
            print(f"Area of contour {i+1}: {area}")
    return frame

更优实现方案（针对规则排列的拼接图） #

如果你的子图是网格状规则排列，用霍夫直线检测分割比轮廓检测更可靠：

def extract_grid_subimages(frame):
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    blur = cv2.GaussianBlur(gray, (5,5), 0)
    edges = cv2.Canny(blur, 50, 150)

    # 检测水平和垂直方向的分割线
    lines = cv2.HoughLinesP(edges, 1, np.pi/180, threshold=100,
                            minLineLength=frame.shape[0]*0.8, maxLineGap=10)
    if lines is None:
        return []

    # 分离水平和垂直线的坐标
    horizontal_lines = []
    vertical_lines = []
    for line in lines:
        x1, y1, x2, y2 = line[0]
        # 判断水平线（y坐标变化极小）
        if abs(y2 - y1) < 10:
            horizontal_lines.append((y1 + y2) // 2)
        # 判断垂直线（x坐标变化极小）
        elif abs(x2 - x1) < 10:
            vertical_lines.append((x1 + x2) // 2)

    # 去重并排序，得到分割线的位置
    horizontal_lines = sorted(list(set(horizontal_lines)))
    vertical_lines = sorted(list(set(vertical_lines)))

    # 根据分割线切割子图
    sub_images = []
    for i in range(len(horizontal_lines)-1):
        for j in range(len(vertical_lines)-1):
            y_start, y_end = horizontal_lines[i], horizontal_lines[i+1]
            x_start, x_end = vertical_lines[j], vertical_lines[j+1]
            # 过滤过小的区域
            if (y_end - y_start) > 20 and (x_end - x_start) > 20:
                sub_img = frame[y_start:y_end, x_start:x_end]
                sub_images.append(sub_img)
    return sub_images

这个方案直接通过分割线定位子图，不会因为单个子图边缘模糊而漏检，适合批量处理规则拼接的图片。

内容的提问来源于stack exchange，提问作者Chaitanya Malhotra