当前位置：首页 > news >正文

【python】OpenCV—Defect Detection

news 2025/8/7 10:31:19

在这里插入图片描述

文章目录

1、功能描述
2、代码实现
3、效果展示
4、完整代码
5、涉及到的库函数
- cv2.bitwise_or()
6、参考

更多有趣的代码示例，可参考【Programming】

1、功能描述

给出一张含有缺陷的样品，识别里面正常的样品和缺陷的样品

2、代码实现

读取图片并可视化

在这里插入图片描述

if __name__ == "__main__":global srcsrc = cv2.imread("1.jpg")  # (800, 640, 3)src_copy = src.copy()cv2.imshow("input", src)

图像转为灰度图并二值化

    # 图像二值化gray = cv2.cvtColor(src, cv2.COLOR_BGR2GRAY)ret, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)

在这里插入图片描述

形态学开运算

    # 形态学开运算se = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3), (-1, -1))binary = cv2.morphologyEx(binary, cv2.MORPH_OPEN, se)cv2.imshow("binary", binary)

在这里插入图片描述

提取轮廓，注意不同版本的 opencv cv2.findContours 返回值数量不一样

    # 轮廓提取# out, contours, hierarchy = cv2.findContours(binary, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)contours, hierarchy = cv2.findContours(binary, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)height, width = src.shape[:2]rects = []

筛选轮廓，轮廓的最小外接矩阵的高大于原图高的二分之一舍弃掉，轮廓面积小于 150 像素的舍弃掉

    # 轮廓筛选for c in range(len(contours)):# 轮廓外接矩形x, y, w, h = cv2.boundingRect(contours[c])# 轮廓面积area = cv2.contourArea(contours[c])if h > (height//2):continueif area < 150:continuerects.append([x, y, w, h])

排序下轮廓，按照纵坐标从小到大排序，并打印出来，轮廓数值表示形式为左上角横纵坐标和宽高

    # 排序轮廓(按照rect中y值排序也就是纵坐标从小到大)rects = sorted(rects, key = lambda x:x[1])for i in range(len(rects)):print(rects[i])

output

[238, 309, 168, 54]
[238, 383, 169, 55]
[239, 457, 168, 56]
[238, 528, 169, 55]

调用 get_template，获取正常样品的模板

    # 获取模板ROItemplate = get_template(binary, rects)cv2.imshow("template", template)

get_template 的实现如下

def get_template(binary, boxes):x, y, w, h = boxes[0]roi = binary[y:y+h, x:x+w]return roi

可以看出，就是把第一个轮廓作为正常样品，也即模板

在这里插入图片描述

把边缘都绘制出来看看

    # 填充边缘for c in range(len(contours)):x, y, w, h = cv2.boundingRect(contours[c])area = cv2.contourArea(contours[c])if h > (height//2):continueif area < 150:continuecv2.drawContours(src_copy, contours, c, (0, 0, 255), 2, 8)# cv2.drawContours(binary, contours, c, (0), 2, 8)cv2.imshow("src_copy", src_copy)

在这里插入图片描述

调用 detect_defect 函数，进行缺陷检测，传入二值化的原图，轮廓坐标，模板图片

    # 检测缺陷defect_boxes = detect_defect(binary, rects, template)

下面具体看看 detect_defect 函数，先遍历每个样品，resize 到和模板一样的大小，cv2.subtract 计算得到样品与模板的差异 mask

形态学、二值化处理下 mask，找到缺陷区域的轮廓，标记成黄色，将缺陷区域叠加到原图（仅在缺陷位置更新）

统计缺陷像素数量 count，如果缺陷像素（二值图里面的白色）高于阈值，则判定该样品是缺陷样品

def detect_defect(binary, boxes, tpl):global srcheight, width = tpl.shapeindex = 1defect_rois = []for x, y, w, h in boxes:# 1. 提取当前ROI并调整大小与模板一致roi = binary[y:y + h, x:x + w]roi = cv2.resize(roi, (width, height))# 2. 计算模板与ROI的差异（掩模）mask = cv2.subtract(tpl, roi)# 3. 形态学开运算去除噪声se = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5), (-1, -1))mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, se)# 4. 二值化处理ret, mask = cv2.threshold(mask, 0, 255, cv2.THRESH_BINARY)# 5. 查找缺陷区域（轮廓）contours, hierarchy = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)# 6. 可视化缺陷区域（黄色标记）color_mask = cv2.cvtColor(mask, cv2.COLOR_GRAY2BGR)cv2.drawContours(color_mask, contours, -1, (0, 255, 255), -1, 8)# 7. 将缺陷区域叠加到原图（仅在缺陷位置更新）mask_roi = src[y:y + height, x:x + width]cv2.bitwise_or(color_mask, mask_roi, mask_roi, mask=mask)# 8. 统计缺陷像素数量count = 0for row in range(height):for col in range(width):pv = mask[row, col]if pv == 255:count += 1# 9. 判断是否存在缺陷if count > 20:defect_rois.append([x, y, w, h])  # 记录缺陷区域else:cv2.rectangle(src, (x, y), (x+w, y+h), (0, 255, 0), 2, 8, 0)  # 标记OKcv2.putText(src, "OK", (x, y - 2), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)cv2.imwrite("mask%d.png"%index, mask)  # 保存中间结果index += 1return defect_rois

看看保存的 mask 细节

样品 1 就是模板，全黑
在这里插入图片描述

样品 2

在这里插入图片描述
样品 3

在这里插入图片描述
样品 4

在这里插入图片描述

绘制缺陷样品，标记黄色字体 NG，框用红色

    for dx, dy, dw, dh in defect_boxes:cv2.rectangle(src, (dx, dy), (dx + dw, dy + dh), (0, 0, 255), 2, 8, 0)cv2.putText(src, "NG", (dx, dy-2), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 255), 2)

在这里插入图片描述

可以看到和样品有差异的区域都被绘制成为了黄色

再绘制一些序号信息 No.1、No.2、No.3、No.4

    index = 1for dx, dy, dw, dh in rects:cv2.putText(src, "No.%d"%index, (dx-60, dy+30), cv2.FONT_HERSHEY_SIMPLEX, .7, (255, 0, 255), 2)index += 1cv2.imshow("result", src)cv2.imwrite("binary2.png", src)cv2.waitKey(0)cv2.destroyAllWindows()

3、效果展示

输入图片

在这里插入图片描述

输出结果

在这里插入图片描述

4、完整代码


import cv2
import numpy as np# 获取模板ROI
def get_template(binary, boxes):x, y, w, h = boxes[0]roi = binary[y:y+h, x:x+w]return roidef detect_defect(binary, boxes, tpl):global srcheight, width = tpl.shapeindex = 1defect_rois = []# 发现缺失for x, y, w, h in boxes:roi = binary[y:y + h, x:x + w]roi = cv2.resize(roi, (width, height))mask = cv2.subtract(tpl, roi)se = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3), (-1, -1))mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, se)ret, mask = cv2.threshold(mask, 0, 255, cv2.THRESH_BINARY)# mask, contours, hierarchy = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)contours, hierarchy = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)color_mask = cv2.cvtColor(mask, cv2.COLOR_GRAY2BGR)cv2.drawContours(color_mask, contours, -1, (0, 255, 255), -1, 8)mask_roi = src[y:y + height, x:x + width]cv2.bitwise_or(color_mask, mask_roi, mask_roi, mask = mask)count = 0for row in range(height):for col in range(width):pv = mask[row, col]if pv == 255:count += 1if count > 20:defect_rois.append([x, y, w, h])else:cv2.rectangle(src, (x, y), (x+w, y+h), (0, 255, 0), 2, 8, 0)cv2.putText(src, "OK", (x, y - 2), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)cv2.imwrite("mask%d.png"%index, mask)index += 1return defect_roisif __name__ == "__main__":global srcsrc = cv2.imread("1.jpg")  # (800, 640, 3)src_copy = src.copy()cv2.imshow("input", src)# 图像二值化gray = cv2.cvtColor(src, cv2.COLOR_BGR2GRAY)ret, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)# 形态学开运算se = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3), (-1, -1))binary = cv2.morphologyEx(binary, cv2.MORPH_OPEN, se)cv2.imshow("binary", binary)# 轮廓提取# out, contours, hierarchy = cv2.findContours(binary, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)contours, hierarchy = cv2.findContours(binary, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)height, width = src.shape[:2]rects = []# 轮廓筛选for c in range(len(contours)):# 轮廓外接矩形x, y, w, h = cv2.boundingRect(contours[c])# 轮廓面积area = cv2.contourArea(contours[c])if h > (height//2):continueif area < 150:continuerects.append([x, y, w, h])# 排序轮廓(按照rect中y值排序也就是纵坐标从小到大)rects = sorted(rects, key = lambda x:x[1])for i in range(len(rects)):print(rects[i])# 获取模板ROItemplate = get_template(binary, rects)cv2.imshow("template", template)# 填充边缘for c in range(len(contours)):x, y, w, h = cv2.boundingRect(contours[c])area = cv2.contourArea(contours[c])if h > (height//2):continueif area < 150:continuecv2.drawContours(src_copy, contours, c, (0, 0, 255), 2, 8)# cv2.drawContours(binary, contours, c, (0), 2, 8)cv2.imshow("src_copy", src_copy)# 检测缺陷defect_boxes = detect_defect(binary, rects, template)for dx, dy, dw, dh in defect_boxes:cv2.rectangle(src, (dx, dy), (dx + dw, dy + dh), (0, 0, 255), 2, 8, 0)cv2.putText(src, "NG", (dx, dy-2), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 255), 2)index = 1for dx, dy, dw, dh in rects:cv2.putText(src, "No.%d"%index, (dx-60, dy+30), cv2.FONT_HERSHEY_SIMPLEX, .7, (255, 0, 255), 2)index += 1cv2.imshow("result", src)cv2.imwrite("binary2.png", src)cv2.waitKey(0)cv2.destroyAllWindows()