OpenCV实现手套表面缺陷检测

---

前言

PVC手套是一款以聚氯乙烯为主原料的手套产品，具有防静电的性能。在许多行业内都会用到，例如电子制造业、药品制造业、化工业、农业等等，运用非常广泛。在PVC手套的生产过程中，会出现有一些质量问题，如粘上油污、蚊虫等情况，这个时候需要将不良品检测出来，以确保供给客户的产品是100%合格。为了检测手套表面缺陷,本文使用OpenCV传统图像处理方法检测出手套表面缺陷。

---

提示：以下是本篇文章正文内容，下面案例可供参考

一、安装所需的包

 pip install virtualenv
 pip install opencv-python
 pip install numpy

二、实现源码解析

1.引入库

代码如下（示例）：

import os
import cv2
import numpy as np

2.代码主干

代码如下（示例）：

IMAGE_SIZE = (500, 500)   #待测图片尺寸，也可以通过路径读取，通过shape获得，方法众多
#如果没有检测到轮廓，THRESHOLD_VALUE需要增加
#如果有不准确的轮廓值需要减少
THRESHOLD_VALUE = 110
MAX_VALUE = 255

#反向阈值
INV_THRESHOLD_VALUE = 50
INV_MAX_VALUE = 255
#Canny
THRESHOLD1 = 100
THRESHOLD2 = 70

#轮廓属性
CON_COLOR = (0, 0, 255)
CON_THICKNESS = 1

#图像叠加特性
WHITE = (255, 255, 255)
BLACK = (0, 0, 0)
GREEN = (0, 255, 0)
RED = (0, 0, 255)
STACK_IMG_SIZE = (200, 200)

while True:
    files = os.listdir('images')
    print("======================================")
    print("=         Available Images           =")
    print("======================================")
    for i in files:
        print('-> {}\t '.format(i), end='')
        if files.index(i) % 3 == 0 and files.index(i) != 0:
            print('\n')
    print("\n======================================")

    # 需要选择具有扩展名的图像名称 (ex: img1.jpeg)
    file = input("Select a file from the directory(q- quit): ").strip()
    if file == 'q' or file == 'Q':
        break
	#检测图像存放位置
    PATH = 'images/' + file
    # 图像路径
    imageOri = cv2.imread(PATH)

    try:
        # 转换为灰度图
        image = cv2.cvtColor(imageOri, cv2.COLOR_BGR2GRAY)
    except:
        print("Invalid Input! Please select correct file(ex: 'imgSample.jpg')")
    else:
        #图像resize
        image = cv2.resize(image, IMAGE_SIZE)
        imageOri = cv2.resize(imageOri, IMAGE_SIZE)
        image = cv2.GaussianBlur(image, (3, 3), 0)

        # 阈值图像，以便您的黑色标记是在白色背景上。
        ret, thresh_basic = cv2.threshold(image, THRESHOLD_VALUE, MAX_VALUE, cv2.THRESH_BINARY)

        #	显示阈值图像 - DEBUGGING
        cv2.imshow("Thresh basic", thresh_basic)

        #thresh_addapt = cv2.adaptiveThreshold(image, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 115, 1)
        # cv2.imshow("Thresh Adapt", thresh_addapt)

        # 取一个大小为5的矩阵作为核
        kernel = np.ones((5, 5), np.uint8)

        #形态学操作——侵蚀前景物体的边界
        #使用形态学来清除多余的标记
        img_erosion = cv2.erode(thresh_basic, kernel, iterations=1)

        # 黑色的标记是黑色背景上的白色然后找到它们的外部轮廓.
        ret, thresh_inv = cv2.threshold(img_erosion, INV_THRESHOLD_VALUE, INV_MAX_VALUE, cv2.THRESH_BINARY_INV)
        #cv2.imshow("INV", thresh_inv)

        # Canny算法找到边缘
        edged = cv2.Canny(img_erosion, THRESHOLD1, THRESHOLD2)
        # show canny edges - DEBUGGING
        #cv2.imshow('Canny', edged)
        #cv2.waitKey(0)

        # 找到轮廓
        # findContours
        contours, hierarchy = cv2.findContours(thresh_inv, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)

        # ++++++++++++++++++
        # -- Image Stack  --
        # ++++++++++++++++++
        font = cv2.FONT_HERSHEY_SIMPLEX

        imageRz = cv2.resize(image, STACK_IMG_SIZE)
        thresh_basicRz = cv2.resize(thresh_basic, STACK_IMG_SIZE)
        img_erosionRz = cv2.resize(img_erosion, STACK_IMG_SIZE)
        thresh_invRz = cv2.resize(thresh_inv, STACK_IMG_SIZE)
        edgedRz = cv2.resize(edged, STACK_IMG_SIZE)

        imageRz = cv2.putText(imageRz, 'GrayScale', (5, 15), font, 0.5, WHITE, 1, cv2.LINE_AA)
        thresh_basicRz = cv2.putText(thresh_basicRz, 'ThresholdBasic', (5, 15), font,
                                     0.5, WHITE, 1,cv2.LINE_AA)
        img_erosionRz = cv2.putText(img_erosionRz, 'Morphology-Erosion', (5, 15), font,
                                    0.5, WHITE, 1, cv2.LINE_AA)
        thresh_invRz = cv2.putText(thresh_invRz, 'Threshold-mode INV', (5, 15), font,
                                   0.5, BLACK, 1, cv2.LINE_AA)
        edgedRz = cv2.putText(edgedRz, 'Canny Edges', (5, 15), font, 0.5, WHITE, 1, cv2.LINE_AA)

        numpy_horizontal_concat = np.concatenate((imageRz, thresh_basicRz, img_erosionRz,
                                                  thresh_invRz, edgedRz), axis=1)

        cv2.imshow('Filtering...', numpy_horizontal_concat)

        # +++++++

        # 得到总轮廓
        num_of_con = str(len(contours) - 1)
        print("Number of Contours found = " + num_of_con)
        if len(contours) > 1:
            print('=       MARKINGS DETECTED            =')

        #显示原始img
        cv2.imshow('Original Image', imageOri)
        # draw contours on original img
        if int(num_of_con) != 0:
            for i in range(int(num_of_con)):
                highlighted_img = cv2.drawContours(imageOri, contours, i, CON_COLOR, CON_THICKNESS)

            highlighted_img = cv2.putText(highlighted_img, 'Approximately {} defect(s) detected'.
                                          format(num_of_con), (5, 15),
                                          font, 0.5, GREEN, 1, cv2.LINE_AA)
        else:
            highlighted_img = cv2.putText(imageOri, 'Unable to detect defects!',
                                          (5, 15), font, 0.5, RED, 2, cv2.LINE_AA)

        # 显示标记的图像
        cv2.imshow('Highlighted Defect', highlighted_img)
        # 保存包含突出显示缺陷的图像
        cv2.imwrite('Output Images/{}_DEFECTS_HIGHLIGHTED.jpg'.format(file.split('.')[0]), highlighted_img)
        cv2.waitKey(0)
        cv2.destroyAllWindows()

---

总结

与人工检测相比，机器视觉检测设备能够高速、高准确性、长期且稳定地进行重复性工作，大量减少出错率，也避免人眼检测弹性疲乏的缺点，提升产品质量及生产速度。