SSIM检测建筑平面图差异的噪声成因及优化方案问询

噪声成因 #

SSIM对地板斑点这类纹理噪声敏感的核心原因是，它的计算基于局部区域的亮度、对比度、结构相似度三个维度。斑点纹理属于高频随机细节差异，即使两张图的宏观建筑结构完全一致，局部像素的微小波动也会拉低SSIM的局部相似度得分，触发阈值检测后被误判为差异区域。

可行优化方法 #

1. 预处理：针对性消除纹理噪声 #

对灰度图先进行滤波处理，优先选择中值滤波（能保留建筑边缘的同时消除斑点/椒盐噪声），也可根据噪声类型选择高斯模糊：

def find_diff(before, after):
    before_grey = cv2.cvtColor(before, cv2.COLOR_BGR2GRAY)
    after_grey = cv2.cvtColor(after, cv2.COLOR_BGR2GRAY)
    
    # 新增：中值滤波消除斑点噪声
    before_grey = cv2.medianBlur(before_grey, 5)  # 核大小选奇数，可根据斑点尺寸调整（3/5/7）
    after_grey = cv2.medianBlur(after_grey, 5)
    
    (score, diff) = compare_ssim(before_grey, after_grey, full=True)
    diff = (diff * 255).astype("uint8")
    print("SSIM: {}".format(score))
    
    thresh = cv2.threshold(diff, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]
    contours = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    contours = imutils.grab_contours(contours)
    
    for c in contours:
        (x, y, w, h) = cv2.boundingRect(c)
        cv2.rectangle(before, (x, y), (x + w, y + h), (0, 0, 255), 2)
        cv2.rectangle(after, (x, y), (x + w, y + h), (0, 0, 255), 2)
    
    return before, after, diff

2. 调整SSIM的计算窗口大小 #

SSIM默认用11x11的窗口计算局部相似度，放大窗口可以忽略局部小纹理的波动，聚焦宏观结构差异：

# 修改compare_ssim调用，指定更大的窗口（必须为奇数）
(score, diff) = compare_ssim(before_grey, after_grey, full=True, win_size=19)

窗口尺寸越大，对局部纹理的容忍度越高，可根据平面图的实际尺寸和纹理密度调整。

3. 后处理：过滤微小误检轮廓 #

斑点噪声对应的轮廓通常面积很小，在绘制差异框前加入面积阈值过滤：

for c in contours:
    # 新增：过滤面积小于阈值的小轮廓
    if cv2.contourArea(c) < 500:  # 阈值根据图像分辨率调整，比如1000对应更大的区域
        continue
    (x, y, w, h) = cv2.boundingRect(c)
    cv2.rectangle(before, (x, y), (x + w, y + h), (0, 0, 255), 2)
    cv2.rectangle(after, (x, y), (x + w, y + h), (0, 0, 255), 2)

4. 替换算法：聚焦结构边缘差异 #

建筑平面图的显著差异（墙体改动、房间布局变化）本质是边缘结构变化，可改用Canny边缘差异检测跳过纹理干扰：

def find_diff(before, after):
    before_grey = cv2.cvtColor(before, cv2.COLOR_BGR2GRAY)
    after_grey = cv2.cvtColor(after, cv2.COLOR_BGR2GRAY)
    
    # 提取边缘
    before_edge = cv2.Canny(before_grey, 50, 150)
    after_edge = cv2.Canny(after_grey, 50, 150)
    
    # 计算边缘差异
    diff = cv2.absdiff(before_edge, after_edge)
    diff = (diff * 255).astype("uint8")
    
    thresh = cv2.threshold(diff, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
    contours = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    contours = imutils.grab_contours(contours)
    
    for c in contours:
        if cv2.contourArea(c) < 500:
            continue
        (x, y, w, h) = cv2.boundingRect(c)
        cv2.rectangle(before, (x, y), (x + w, y + h), (0, 0, 255), 2)
        cv2.rectangle(after, (x, y), (x + w, y + h), (0, 0, 255), 2)
    
    return before, after, diff

这种方法直接聚焦结构边缘的变化，完全避开地板斑点这类纹理噪声的干扰。

内容的提问来源于stack exchange，提问作者bachtick