express_label_identification_python
/
express.py

import cv2 as cv
import numpy as np

###进行泛洪填充，去除图片的背景
def ClearBackground(img):
    h, w = img.shape[:2]
    # 低通滤波，进行降噪
    blured = cv.blur(img, (5, 5))
    # cv.namedWindow("blured",0)
    # cv.imshow("blured",blured)
    # 掩码长和宽都比输入图像多两个像素点，满水填充不会超出掩码的非零边缘
    mask = np.zeros((h + 2, w + 2), np.uint8)
    # 进行泛洪填充,去除图片的背景
    cv.floodFill(blured, mask, (w - 1, h - 1), (255, 255, 255), (2, 2, 2), (3, 3, 3), 8)
    cv.namedWindow("ClearBack", 0)
    cv.imshow("ClearBack", blured)
    return blured

###做闭运算，使图形的缝隙连接，形成不同的区域，返回处理后的二值图
def CloseImage(img,w,h):
    # 转为灰度图像
    gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
    # 使用Scharr滤波器操作（指定使用ksize = -1）构造灰度图在水平和竖直方向上的梯度幅值表示
    gradX = cv.Sobel(gray, ddepth=cv.CV_32F, dx=1, dy=0, ksize=-1)
    gradY = cv.Sobel(gray, ddepth=cv.CV_32F, dx=0, dy=1, ksize=-1)
    # Scharr操作之后，从x-gradient中减去y-gradient，通过减法操作，得到高水平梯度和低竖直梯度的图像区域
    gradient = cv.subtract(gradX, gradY)
    gradient = cv.convertScaleAbs(gradient)
    # 梯度图进行平均模糊，有助于平滑梯度表征的图形中的高频噪声
    blurred = cv.blur(gradient, (3, 3))
    # 图像二值化
    (_, thresh) = cv.threshold(blurred, 100, 255, cv.THRESH_BINARY)
    # cv2.getStructuringElement构造长方形。宽度大于长度，可以消除条形码中各垂直条之间的缝隙
    kernel = cv.getStructuringElement(cv.MORPH_RECT, (w, h))
    # 形态学闭运算，消除缝隙
    closed = cv.morphologyEx(thresh, cv.MORPH_CLOSE, kernel)
    # 腐蚀，膨胀，去除其中的小噪点
    closed = cv.erode(closed, None, iterations=4)
    closed = cv.dilate(closed, None, iterations=4)
    return closed

###得到条形码+收寄信息的区域
# 传入进过了闭运算的二值图
# 返回一个box
def GetRegion(img):
    h, w = img.shape[:2]
    # 寻找图像的轮廓
    cnts = cv.findContours(img, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
    cnts = cnts[1]
    # 对所有轮廓按面积降序
    cc = sorted(cnts, key=cv.contourArea, reverse=True)
    ###确定二维码区域为哪一个
    # 条形码区域为面积最大的
    cNum = cc[0]
    cConnect = cc[1]
    # 得到轮廓的矩阵信息
    rect = cv.minAreaRect(cNum)
    box = np.int0(cv.boxPoints(rect))
    rectC = cv.minAreaRect(cConnect)
    boxC = np.int0(cv.boxPoints(rectC))
    mi = 99999
    ma = -1
    for i in range(0, 4):
        x1 = int(box[i][0])
        if x1 > ma:
            ma = x1
        if x1 < mi:
            mi = x1
    # 如果占比大于图片宽度的0.7，则不是条形码区域
    # print((ma - mi) / w)
    if (ma - mi) / w > 0.7:
        if len(cc) > 2:
            # 条形码区域为面积第二大的
            cNum = cc[1]
            cConnect = cc[0]
        else:
            print("图片处理存在错误！错误地点1:express.GetRegion()")
    rect = cv.minAreaRect(cNum)
    box = np.int0(cv.boxPoints(rect))
    rectC = cv.minAreaRect(cConnect)
    boxC = np.int0(cv.boxPoints(rectC))
    # print(box)
    return box,boxC

###根据已有box裁剪出快递单号的图片
def CutExpressImage(box,img):
    # 按最后一列排序
    box1 = box[np.lexsort(box.T)]
    # 确定数字区域的宽高与起点
    roiw = abs(box1[0][0] - box1[1][0])
    # 数字区域 取条形码的2/3高度
    roih = int(2 * abs(box1[0][1] - box1[2][1]) / 3)
    # 选取条形码下方的数字区域
    # 确定左上角的点，以及宽高，裁剪
    roiPoint = box1[2]
    if box[2][0] > box1[3][0]:
        roiPoint = box1[3]
    # print(roiw, " ", roih, " ", roiPoint)
    # 裁剪
    diget_img = img[roiPoint[1]:(roiPoint[1] + roih), roiPoint[0]:(roiPoint[0] + roiw)]
    # 保存数字区域
    cv.imwrite("./img/expressImg.jpg",diget_img)
    cv.namedWindow("ImageExpress",0)
    cv.imshow("ImageExpress", diget_img)

###根据已有box裁剪出联系人信息的图片
def CutConnectImage(boxC,img):
    ###确定包含信息的区域
    # 按最后一列排序
    boxC1 = boxC[np.lexsort(boxC.T)]
    # 确定宽高与起点
    roiCw = abs(boxC1[0][0] - boxC1[1][0])
    # 高度
    roiCh = abs(boxC1[0][1] - boxC1[2][1])
    # 选取条形码下方的数字区域
    # 确定左上角的点，以及宽高，裁剪
    roiCPoint = boxC1[0]
    if boxC[0][0] > boxC1[1][0]:
        roiCPoint = boxC1[1]
    # 裁剪信息区域
    connect_img = img[roiCPoint[1]:(roiCPoint[1] + roiCh), roiCPoint[0]:(roiCPoint[0] + roiCw)]
    cv.imwrite("img/connectImg.jpg", connect_img)
    cv.namedWindow("ImageConnect", 0)
    cv.imshow("ImageConnect", connect_img)

###展现已经得到的ROI区域
def ShowImageROI(blured,closed,box,boxC):
    # 在图片上画出快递单号区域，显示为红色
    cv.drawContours(blured, [box], -1, (0, 0, 255), 8)
    # 在图片上画出信息区域，显示为绿色
    cv.drawContours(blured, [boxC], -1, (0, 255, 0), 8)
    cv.namedWindow("ImageROI", 0)
    cv.imshow("ImageROI", blured)
    cv.namedWindow("ImageClosed", 0)
    cv.imshow("ImageClosed", closed)
    cv.waitKey(0)

###读入图片，快递识别主程序
def GetImageROI(path):
    #读入图像
    orginImg = cv.imread(path) #"img/c2.jpg"
    ###去除背景
    blured = ClearBackground(orginImg)
    ###闭运算，消除同一区域的缝隙
    closed = CloseImage(blured,80,3)
    ###确定条形码+信息的区域
    box, boxC = GetRegion(closed)
    ###将快递单号裁剪并保存
    CutExpressImage(box,blured)
    ###将联系电话信息裁剪并保存
    CutConnectImage(boxC,blured)
    ###为便于观察，显示出ROI区域
    ShowImageROI(blured,closed,box,boxC)

if __name__ == '__main__':
    # ImageRead("img/c1.jpg")#
    # ImageRead("img/c2.jpg")#
    # ImageRead("img/c3.jpg")
    # ImageRead("img/c4.jpg")
    # ImageRead("img/c5.jpg")#
    # ImageRead("img/c6.jpg")
    # ImageRead("img/z1.jpg")
    GetImageROI("img/z2.jpg")
    # ImageRead("img/z3.jpg")#
    # ImageRead("img/z4.jpg")#
    # ImageRead("img/z5.jpg")
    # ImageRead("img/z6.jpg")
#得到灰度图
# grayImg = cv.cvtColor(orginImg,cv.COLOR_BGR2GRAY)
###
# #定义结构元素
# kernel = cv.getStructuringElement(cv.MORPH_RECT,(50, 50))
# #开闭运算，先开运算去除背景噪声，再继续闭运算填充目标内的孔洞
# opened = cv.morphologyEx(grayImg, cv.MORPH_OPEN, kernel)
# closed = cv.morphologyEx(opened, cv.MORPH_CLOSE, kernel)
# cv.imshow("closed", closed)
# #求二值图
# ret, binary = cv.threshold(closed,250,255,cv.THRESH_BINARY)
# cv.imshow("binary", binary)
# #找到轮廓
# _,contours, hierarchy = cv.findContours(binary,cv.RETR_TREE,cv.CHAIN_APPROX_SIMPLE)
# #绘制轮廓
# cv.drawContours(orginImg,contours,-1,(0,0,255),3)
# #绘制结果
# cv.namedWindow("result",0)
# cv.imshow("result", orginImg)
# #自适应阈值
# binaryImg = cv.adaptiveThreshold(grayImg, 255, cv.ADAPTIVE_THRESH_MEAN_C, cv.THRESH_BINARY, 11, 2)
# cannyImg = cv.Canny(blured, 50, 150)
# # 寻找出所有轮廓集合 http://blog.sina.com.cn/s/blog_662c78590100z0rg.html
# contours = cv.findContours(cannyImg, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)

# contours = contours[1]
# # 按面积降序
# contours = sorted(contours, key=cv.contourArea, reverse=True)
# 曲线逼近
# 简化边界-方便找出轮廓
# cv.drawContours(orginImg, approx, -1, (0, 0, 255), 30)

# #自适应阈值
# binaryWImg = cv.adaptiveThreshold(binaryImg, 255, cv.ADAPTIVE_THRESH_MEAN_C, cv.THRESH_BINARY, 101, 2)
# #进行均值滤波
# meanImg = cv.blur(binaryWImg, (5, 5))
# #进行固定阈值过滤
# meanImg = cv.threshold(meanImg, 100, 255, cv.THRESH_BINARY)[1]
# #对图片进行腐蚀膨胀3次操作，让选择题突出，便于处理
# meanImg = cv.erode(meanImg, None, iterations=3)
# #进行均值滤波
# meanImg = cv.adaptiveThreshold(meanImg, 255, cv.ADAPTIVE_THRESH_MEAN_C, cv.THRESH_BINARY, 101, 2)
# meanImg = cv.blur(meanImg, (25, 25))
# meanImg = cv.threshold(meanImg, 100, 255, cv.THRESH_BINARY)[1]

# cv.namedWindow("GrayImage",0)
# cv.imshow("GrayImage",cannyImg)
# cv.waitKey(0)

#1.预处理
#2.ROI定位条形码区域
#3.裁剪数字区域