1.提取绿色区域

1.读取图片，进行高斯模糊，并转化为hsv图片，这是为了方便颜色的提取。

frame = cv2.imread(r"C:\Users\xjy\Desktop\hanjia\1.jpg")
gs_frame = cv2.GaussianBlur(frame, (5, 5), 0)  # 高斯模糊
hsv = cv2.cvtColor(gs_frame, cv2.COLOR_BGR2HSV)  # 转化成HSV图像

2.进行腐蚀减少干扰，去除背景，转化为二值化图片

ball_color = 'green'
#三种颜色在hsv模式下的范围
color_dist = {'red': {'Lower': np.array([0, 60, 60]), 'Upper': np.array([6, 255, 255])},
              'blue': {'Lower': np.array([100, 80, 46]), 'Upper': np.array([124, 255, 255])},
              'green': {'Lower': np.array([35, 43, 35]), 'Upper': np.array([90, 255, 255])},
              }

erode_hsv = cv2.erode(hsv, None, iterations=2)  # 腐蚀 
inRange_hsv = cv2.inRange(erode_hsv, color_dist[ball_color]['Lower'], color_dist[ball_color]['Upper'])

3.获取矩形的轮廓，提取最小外接矩形的坐标，在原图上裁剪出来

inRange_hsv = cv2.inRange(erode_hsv, color_dist[ball_color]['Lower'], color_dist[ball_color]['Upper'])
cnts = cv2.findContours(inRange_hsv.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[-2]#获取图片中的轮廓
c = max(cnts, key=cv2.contourArea) #在边界中找出面积最大的区域
rect = cv2.minAreaRect(c)    # 绘制出该区域的最小外接矩形
box = cv2.boxPoints(rect)   # 记录该矩形四个点的位置坐标
box = np.int0(box)   #将坐标转化为整数

x, y, w, h = cv2.boundingRect(box) #  获取最小外接轴对齐矩形的坐标

image = frame[y:y + h, x:x + w]  #获取roi区域

2.对剪切后的图片进行处理

1.转为二值化图片，进行膨胀

# 灰度图像
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# 二值化
ret, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)
#变成白底黑字
cv2.bitwise_not(binary, binary)
#进行膨胀运算，方便后期轮廓检测
kernel = np.ones((5, 5), np.uint8)
bin2 = cv2.erode(binary, kernel)
cv2.imshow("pz",bin2)
cv2.waitKey(0)

2.对图片中较小的轮廓进行填充，减小干扰

# 小轮廓去除
contours, hierarchy = cv2.findContours(bin2, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
fill = []
for contour in contours:
    #计算轮廓周长
    area = cv2.contourArea((contour))
    #如果周长小于一个值就添加进列表
    if area < 250:
        fill.append(contour)
#将图中轮廓周长小的部位填涂
bin2 = cv2.fillPoly(bin2, fill, (255, 255, 255))

3.选取一定面积大小的轮廓

# 轮廓检测
contours, hierarchy = cv2.findContours(bin2, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
draw_img2 = bin2.copy()
# 筛选轮廓：
need_contours = []
for contour in contours:
    #计算轮廓的左上角坐标xy和宽高wh
    x, y, w, h = cv2.boundingRect(contour)
    #计算矩形面积
    rectangle_area = w * h
    #一定大小的面积才圈出轮廓
    if (rectangle_area < 10000 and rectangle_area > 600):
        need_contours.append(contour)

res = cv2.drawContours(draw_img2, need_contours, -1, (255, 0, 0), 2)
cv2.imshow("res", res)
cv2.waitKey(0)

4.对剩余轮廓画矩形

draw_img2 = image.copy()
for contour in need_contours:
    # 计算轮廓的左上角坐标xy和宽高wh
    x, y, w, h = cv2.boundingRect(contour)
    #记录数字的轮廓坐标
    xy.append([x,y,x+w,y+h])
    #在每个轮廓外花矩形框
    draw_img2 = cv2.rectangle(draw_img2, (x, y), (x + w, y + h), (0, 0, 255), 2)

cv2.imshow("img3", draw_img2)

3.剪切数字

    #根据x的值比较，x之间差大于100的图片为竖图，需要根据y来排序
    if abs(xy[0][0]-xy[-1][0] )>100:
        xy=sorted(xy, key=lambda x: x[0])
    else:
        xy=sorted(xy, key=lambda x: x[1])
    #保存每个数字的图片
    for i in range(len(xy)):
        cat = bin2[xy[i][1]:xy[i][3],xy[i][0]:xy[i][2]]
        size = (40, 30)
        cat = cv2.resize(cat, size)
        cv2.imwrite(r"./cat/{}.jpg".format(i+1), cat)

4.读取模板图片

# 模板文件夹路径
path = './moban'
templates = []
template_ids = []
filelist = os.listdir(path)
for files in filelist:
    Olddir = os.path.join(path, files)
    # 如果是文件夹
    if os.path.isdir(Olddir):
        # 遍历小文件夹
        filelist = os.listdir(Olddir)
        for files in filelist:
            Olddir2 = os.path.join(Olddir, files)
            templates.append(cv2.imread(Olddir2, 0))
            a = files.find("_")
            template_ids.append(files[0:a])

5.进行模板匹配并输出结果

# 遍历模板列表
list2=[]
i=1
flag=False
while i<=len(xy):
    img_path=r"C:\Users\xjy\Desktop\hanjia\cat\{}.jpg".format(i)
    frame = cv2.imread(img_path)
    #灰度图
    frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    i+=1
    for template, template_id in zip(templates, template_ids):
        #进行模板匹配
        res = cv2.matchTemplate(frame, template, cv2.TM_CCOEFF_NORMED)
        #设置匹配度
        threshold = 0.8
        loc = np.where(res >= threshold)
        # 为每个匹配到的位置添加信息到结果字典
        for pt in zip(*loc[::-1]):
            if template_id=="10":
                list2.append("/")
            else:
                list2.append(template_id)
            flag=True
            break
        if flag==True:
            flag=False
            break
result = ''.join(list2)
print(result)

ocr识别

下载ocr识别

https://digi.bib.uni-mannheim.de/tesseract/

下载完后在终端下载pytesseract库，并修改参数，地址为你安装的地址

1.识别

test_message = Image.fromarray(bin2)
text = pytesseract.image_to_string(test_message,lang='eng')
print(f'识别结果：{text}')

结果：