opencv-python实际演练(一）图像识别（1）目标区域提取

背景

在现场下军棋时需要三个人，其中一个人当裁判。如果只有两个人，又想玩军棋，就需要有一个自动裁判机制。想通过手机自动识别棋子上的文字，目前还没有看到专门实现这个功能的软件，因此想自已动手试一试。

准备工作

用手机拍摄了一张上面有两个棋子的图片（模拟生成一副图片后再用手机对着屏幕拍摄的，以后再对着真实的棋子拍照吧）

在网上查到了一篇参考文献《基于python+opencv的图像目标区域自动提取（本项目为提取纸张中的内容）》
参考文中的代码，并做了一些修改。对原作者的无私分享深表感谢！

提取结果

运行后的实际效果是得到两个图片文件

python代码

#coding:utf-8
#用手机拍摄的两个军棋棋子照片，将这两个棋子的内容从照片中提取出来，供下一步的文本识别使用
import cv2
import numpy as np
import math

#配置数据
class Config:
    def __init__(self):
        pass
    src = "photo1.jpg"
    resizeRate = 0.5
    min_area = 5000
    min_contours = 8
    threshold_thresh = 50
    epsilon_start = 10
    epsilon_step = 10

''' 对坐标点进行排序 @return [top-left, top-right, bottom-right, bottom-left] '''
def order_points(pts):
    # initialzie a list of coordinates that will be ordered
    # such that the first entry in the list is the top-left,
    # the second entry is the top-right, the third is the
    # bottom-right, and the fourth is the bottom-left
    rect = np.zeros((4, 2), dtype="float32")

    # the top-left point will have the smallest sum, whereas
    # the bottom-right point will have the largest sum
    s = pts.sum(axis=1)
    rect[0] = pts[np.argmin(s)]
    rect[2] = pts[np.argmax(s)]

    # now, compute the difference between the points, the
    # top-right point will have the smallest difference,
    # whereas the bottom-left will have the largest difference
    diff = np.diff(pts, axis=1)
    rect[1] = pts[np.argmin(diff)]
    rect[3] = pts[np.argmax(diff)]

    # return the ordered coordinates
    return rect

# 求两点间的距离
def point_distance(a,b):
    return int(np.sqrt(np.sum(np.square(a - b))))

# 找出外接四边形, c是轮廓的坐标数组
def boundingBox(idx,c):
    if len(c) < Config.min_contours: 
        return None
    epsilon = Config.epsilon_start
    while True:
        approxBox = cv2.approxPolyDP(c,epsilon,True)
        #求出拟合得到的多边形的面积
        theArea = math.fabs(cv2.contourArea(approxBox))
        #输出拟合信息
        print("contour idx: %d ,contour_len: %d ,epsilon: %d ,approx_len: %d ,approx_area: %s"%(idx,len(c),epsilon,len(approxBox),theArea))
        if (len(approxBox) < 4):
            return None
        if theArea > Config.min_area:
            if (len(approxBox) > 4):
                # epsilon 增长一个步长值
                epsilon += Config.epsilon_step               
                continue
            else: #approx的长度为4，表明已经拟合成矩形了 
                #转换成4*2的数组
                approxBox = approxBox.reshape((4, 2))                
                return approxBox                
        else:
            print("failed to find boundingBox,idx = %d area=%f"%(idx, theArea))
            return None

#-----------------------------------------------------------------------------------------------

# 开始图像处理，读取图片文件
image = cv2.imread(Config.src)
#print(image.shape)

#获取原始图像的大小
srcHeight,srcWidth ,channels = image.shape

#对原始图像进行缩放
image= cv2.resize(image,(int(srcWidth*Config.resizeRate),int(srcHeight*Config.resizeRate))) 
#cv2.imshow("image", image)

#转成灰度图
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) 

# 中值滤波平滑，消除噪声
# 当图片缩小后，中值滤波的孔径也要相应的缩小，否则会将有效的轮廓擦除
#binary = cv2.medianBlur(gray,7)
binary = cv2.medianBlur(gray,3)  

#转换为二值图像
ret, binary = cv2.threshold(binary, Config.threshold_thresh, 255, cv2.THRESH_BINARY)
#显示转换后的二值图像
cv2.imshow("binary", binary)

# 进行2次腐蚀操作（erosion）
# 腐蚀操作将会腐蚀图像中白色像素，可以将断开的线段连接起来
binary = cv2.erode (binary, None, iterations = 2)
#显示腐蚀后的图像
cv2.imshow("erode", binary)

# canny 边缘检测
binary = cv2.Canny(binary, 0, 60, apertureSize = 3)
#显示边缘检测的结果
cv2.imshow("Canny", binary)

# 提取轮廓
contours,_ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
# 输出轮廓数目
print("the count of contours is %d \n"%(len(contours)))

#针对每个轮廓，拟合外接四边形,如果成功，则将该区域切割出来，作透视变换，并保存为图片文件
for idx,c in enumerate(contours):
    approxBox = boundingBox(idx,c)
    if approxBox is None: 
        print("\n")
        continue
    
     # 获取最小矩形包络
    rect = cv2.minAreaRect(approxBox)
    box = cv2.boxPoints(rect)
    box = np.int0(box)
    box = box.reshape(4,2)
    box = order_points(box)
    print("boundingBox：\n",box)   

    # 待切割区域的原始位置，
    # approxPolygon 点重排序, [top-left, top-right, bottom-right, bottom-left]
    src_rect = order_points(approxBox)  
    print("src_rect：\n",src_rect)
   
    w,h = point_distance(box[0],box[1]), point_distance(box[1],box[2])
    print("w = %d ,h= %d "%(w,h))
    
    # 生成透视变换矩阵
    dst_rect = np.array([
        [0, 0],
        [w - 1, 0],
        [w - 1, h - 1],
        [0, h - 1]],
        dtype="float32")

    # 透视变换
    M = cv2.getPerspectiveTransform(src_rect, dst_rect)

    #得到透视变换后的图像
    warped = cv2.warpPerspective(image, M, (w, h))

    #将变换后的结果图像写入png文件
    cv2.imwrite("output/piece%d.png"%idx, warped, [int(cv2.IMWRITE_PNG_COMPRESSION), 9])

    print("\n")

                
print('over')
cv2.waitKey(0)

运行结果如下：

弹出了三个图像窗口
1.二值化后的结果

2 腐蚀后的结果，将断开的线段连起来

3 canny 边缘检测后的结果

最后在output目录下生成了前面展示的两个结果文件

补充说明，以上代码要正常运行的前提是 numpy 与opencv-python模块已安装好。不过这两个模块的安装都很简单，用pip install numpy 与 pip install opencv-python就OK了