#coding='utf-8' import cv2 import numpy as np import random import math def draw_contours(imgColor,minArea=50): if len(imgColor.shape)>2: imgResult=cv2.cvtColor(imgColor,cv2.COLOR_BGR2GRAY) else: imgResult = imgColor.copy() ret, img_binary = cv2.threshold(imgResult, 125, 255, 0) contours0, hierarchy = cv2.findContours(img_binary, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) contours=list([]) for cnt in contours0: rect = cv2.minAreaRect(cnt) if rect[1][0] * rect[1][1] < minArea: # 此矩形区域的面积小于10，则忽略 continue contours.append(cnt)#contours=contours.append(cnt)会返回NoneType错误 cv2.drawContours(imgResult, contours, -1, (0, 0, 255), 3) return imgResult def getDis(pointX, pointY, lineX1, lineY1, lineX2, lineY2): '''返回点线之间的距离 ''' a = lineY2 - lineY1 b = lineX1 - lineX2 c = lineX2 * lineY1 - lineX1 * lineY2 dis = (math.fabs(a * pointX + b * pointY + c)) / (math.pow(a * a + b * b, 0.5)) return dis def get_region_description(imgColor,minArea=50): if len(imgColor.shape)>2: imgResult=cv2.cvtColor(imgColor,cv2.COLOR_BGR2GRAY) else: imgResult = imgColor.copy() _, imgBinary = cv2.threshold(imgResult, 125, 255, 0) row,col=imgBinary.shape[:2] contours0, hierarchy = cv2.findContours(imgBinary, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE) # 去除小面积区域 contours_=[] for cnt in contours0: rect = cv2.minAreaRect(cnt) if rect[1][0] * rect[1][1] < minArea: # 此矩形区域的面积小于10，则忽略 continue contours_.append(cnt)#contours=contours.append(cnt)会返回NoneType错误 cv2.drawContours(imgResult, contours_, -1, (0, 0, 255), 3) contours=[] #存放区域描述子，包括重心、周长、面积、圆形度、矩形度 # 进行区域描述 for cnt in contours_: oneContour = [] # 存放一个区域的描述子 # minAreaRect返回值：（最小外接矩形的中心（x，y），（宽度，高度），旋转角度） #角度是【-90，0】，0度会写成-90度，方向不准确 rect = cv2.minAreaRect(cnt) # maxRegion=min(rect[1][0],rect[1][1])/2 # maxV=maxRegion*maxRegion #区域最大半径平方值 points=cv2.boxPoints(rect)#可以返回四个点的值 #获得重心 cx,cy=int(rect[0][0]),int(rect[0][1]) oneContour.append((cx,cy)) #添加重心 # #获得区域方向，并绘制方向 '''minV = math.inf #区域上的点到直线距离和的最小值 min_i = 0 min_j = 0 for i in range(row): for j in range(col): sum = 0 # 获得一个区域最大半径圆上的点，做点到质心的线 if (i - cx) * (i - cx) + (j - cy) * (j - cy) == maxV: # 一个圆上的点 for c0 in cnt[:][:]: # cnt中存放中区域轮廓上的点坐标 sum += getDis(c0[0][0], c0[0][1], cx, cy, i, j) if sum < minV: minV = sum min_i = i min_j = j cv2.line(imgResult, (cx, cy), (min_i, min_j), (0, 0, 0), 3, 8) regionDirect = math.atan2(min_j-cy,min_i-cx) # regionDirect = rect[2] oneContour.append(regionDirect)''' minV = math.inf findPoint=(0,0) rectPoints=[]#获得区域最上边、下边、左边、右边线上的点,做点到中心的线 minRow=int(min(points[i][0] for i in range(4))) #最小行坐标 maxRow=int(max(points[i][0] for i in range(4))) minCol=int(min(points[i][1] for i in range(4))) maxCol=int(max(points[i][1] for i in range(4))) highPointNum=int(maxRow-minRow+1) widthPointNum=int(maxCol-minCol+1) fixI1=zip([minRow]*widthPointNum,range(minCol,maxCol+1,1)) rectPoints.extend(fixI1) fixI2=zip([maxRow]*widthPointNum,range(minCol,maxCol+1,1)) rectPoints.extend(fixI2) fixJ1=zip(range(minRow,maxRow+1,1),[minCol]*highPointNum) rectPoints.extend(fixJ1) fixJ2=zip(range(minRow,maxRow+1,1),[maxCol]*highPointNum) rectPoints.extend(fixJ2) for point in rectPoints: sum = 0 for c0 in cnt[:][:]: # cnt中存放中区域轮廓上的点坐标 sum += getDis(c0[0][0], c0[0][1], cx, cy, point[0],point[1]) if sum < minV: minV = sum findPoint=point cv2.line(imgResult, (cx, cy), findPoint, (0, 0, 0), 3, 8) regionDirect =math.atan2(cy-findPoint[1],cx-findPoint[0])*180/3.14 oneContour.append(regionDirect) # 添加重心 # 计算轮廓面积和周长 area = cv2.contourArea(cnt) perimeter = cv2.arcLength(cnt,False) circle = 4.0*np.pi*area/(perimeter*perimeter) # 计算圆形度 rectangle = area/(rect[1][0]*rect[1][1]) # 计算矩形度 oneContour.append(area) oneContour.append(perimeter) oneContour.append(circle) oneContour.append(rectangle) contours.append(oneContour) return imgResult,contours # Hu Moments 胡不变矩 wiki： https://en.wikipedia.org/wiki/Image_moment # 公式 https://docs.opencv.org/4.1.2/d3/dc0/group__imgproc__shape.html#gab001db45c1f1af6cbdbe64df04c4e944 # 代码 https://github.com/opencv/opencv/blob/b6a58818bb6b30a1f9d982b3f3f53228ea5a13c1/modules/imgproc/src/moments.cpp # void cv::HuMoments( const Moments& m, double hu[7] ) # { # CV_INSTRUMENT_REGION(); # # double t0 = m.nu30 + m.nu12; # double t1 = m.nu21 + m.nu03; # # double q0 = t0 * t0, q1 = t1 * t1; # # double n4 = 4 * m.nu11; # double s = m.nu20 + m.nu02; # double d = m.nu20 - m.nu02; # # hu[0] = s; # hu[1] = d * d + n4 * m.nu11; # hu[3] = q0 + q1; # hu[5] = d * (q0 - q1) + n4 * t0 * t1; # # t0 *= q0 - 3 * q1; # t1 *= 3 * q0 - q1; # # q0 = m.nu30 - 3 * m.nu12; # q1 = 3 * m.nu21 - m.nu03; # # hu[2] = q0 * q0 + q1 * q1; # hu[4] = q0 * t0 + q1 * t1; # hu[6] = q1 * t0 - q0 * t1; # } #求图像的24个矩 def get_moments(img,moments=None): # row == heigh == Point.y # col == width == Point.x # Mat::at(Point(x, y)) == Mat::at(y,x) # https://blog.csdn.net/puqian13/article/details/87937483 mom = [0] * 10 #10个原点矩 umom = [0] * 7 #7个中心矩 numom = [0] * 7 #7个归一化中心矩 rows, cols = img.shape for y in range(rows): # temp var x0 = 0 x1 = 0 x2 = 0 x3 = 0 for x in range(cols): p = img[y,x] xp = x * p xxp = x * xp x0 = x0 + p x1 = x1 + xp x2 = x2 + xxp x3 = x3 + xxp * x py = y * x0 sy = y*y mom[9] += (py) * sy # m03 mom[8] += (x1) * sy # m12 mom[7] += (x2) * y # m21 mom[6] += x3 # m30 mom[5] += x0 * sy # m02 mom[4] += x1 * y # m11 mom[3] += x2 # m20 mom[2] += py # m01 mom[1] += x1 # m10 mom[0] += x0 # m00 x_a = mom[1] / mom[0] y_a = mom[2] / mom[0] for y in range(0,rows): x_a_0 = 0 x_a_1 = 0 x_a_2 = 0 x_a_3 = 0 for x in range(0,cols): p = img[y,x] x_a_0 = x_a_0 + p x_a_1 += p * (x - x_a ) x_a_2 += x_a_1 * (x - x_a ) x_a_3 += x_a_2 * (x - x_a ) y_a_1 = (y - y_a) y_a_2 = y_a_1 * (y - y_a) y_a_3 = y_a_2 * (y - y_a) umom[0] += x_a_2 umom[1] += x_a_1 * y_a_1 umom[2] += x_a_0 * y_a_2 umom[3] += x_a_3 umom[4] += x_a_2 * y_a_1 umom[5] += x_a_1 * y_a_2 umom[6] += x_a_0 * y_a_3 cx = mom[1] * 1.0 / mom[0] cy = mom[2] * 1.0 / mom[0] umom[0] = mom[3] - mom[1] * cx umom[1] = mom[4] - mom[1] * cy umom[2] = mom[5] - mom[2] * cy umom[3] = mom[6] - cx * (3 * umom[0] + cx * mom[1]) umom[4] = mom[7] - cx * (2 * umom[1] + cx * mom[2]) - cy * umom[0] umom[5] = mom[8] - cy * (2 * umom[1] + cy * mom[1]) - cx * umom[2] umom[6] = mom[9] - cy * (3 * umom[2] + cy * mom[2]) # nu inv_sqrt_m00 = np.sqrt(abs(1.0 / mom[0])) s2 = (1.0 / mom[0]) * (1.0 / mom[0]) s3 = s2 * inv_sqrt_m00 numom[0] = umom[0] * s2 numom[1] = umom[1] * s2 numom[2] = umom[2] * s2 numom[3] = umom[3] * s3 numom[4] = umom[4] * s3 numom[5] = umom[5] * s3 numom[6] = umom[6] * s3 moments = mom + umom + numom return moments def get_hu_moments(numom): ''' :param numom: 图像的原点矩 :return: 返回7个不变矩 ''' hu = [0]*7 t0 = numom[3] + numom[5] t1 = numom[4] + numom[6] q0 = t0 * t0 q1 = t1 * t1 n4 = 4 * numom[1]; s = numom[0] + numom[2]; d = numom[0] - numom[2]; hu[0] = s hu[1] = d * d + n4 * numom[1] hu[3] = q0 + q1 hu[5] = d * (q0 - q1) + n4 * t0 * t1 t0 *= q0 - 3 * q1 t1 *= 3 * q0 - q1 q0 = numom[3] - 3 * numom[5] q1 = 3 * numom[4] - numom[6] ; hu[2] = q0 * q0 + q1 * q1 hu[4] = q0 * t0 + q1 * t1 hu[6] = q1 * t0 - q0 * t1 return hu def moment_invariants(img): textLists=[] #用来存放几何变换名称 imgLists=[] #用来存放几何变换图像 # 旋转45度 rotate_45_matrix = cv2.getRotationMatrix2D((img.shape[1] // 2, img.shape[0] // 2), -45, 1) rotate_45_image = cv2.warpAffine(img, rotate_45_matrix, dsize=(img.shape[1], img.shape[0])) textLists.append("rotate-45") imgLists.append(rotate_45_image) # 平移 M = np.float32([[1, 0, 50], [0, 1, 50]]) translation_img = cv2.warpAffine(img, M, dsize=(img.shape[1], img.shape[0])) textLists.append("translation") imgLists.append(translation_img) # 顺时针旋转90 rotate_90_iamge = cv2.rotate(img, rotateCode=cv2.ROTATE_90_CLOCKWISE) textLists.append("rotate-90") imgLists.append(rotate_90_iamge) # 顺时针旋转180 rotate_180_iamge = cv2.rotate(img, cv2.ROTATE_180) textLists.append("rotate-180") imgLists.append(rotate_180_iamge) # 缩小一半 resize_0_5_img=np.zeros(img.shape) halfImg = cv2.resize(img, (img.shape[1] // 2, img.shape[0] // 2)) textLists.append("shrink ") imgLists.append(halfImg) # 逆时针旋转90 rotate_270_image = cv2.rotate(img, rotateCode=cv2.ROTATE_90_COUNTERCLOCKWISE) textLists.append("rotate-270") imgLists.append(rotate_270_image) # calc the moments of the image '''get 24 moments 10 spatial moments 空间矩、原点矩 m00 m10 m01 m20 m11 m02 m30 m21 m12 m03 7 central moments 七个中心矩 mu20 mu11 mu02 mu30 mu21 mu12 mu03 7 central normalized moments个归一化中心矩 nu20 nu11 nu02 nu30 nu21 nu12 nu03 本来应该都是10个的 but mu00 = m00, nu00 = 1 nu10 = mu10 = mu01 = mu10 = 0, hence the values are not stored. 原理 见论文 论文 https://pdfs.semanticscholar.org/afc2/e9d5dfbd666bf4dd34adeb78a17393c8ee64.pdf?_ga=2.259665167.462545856.1577780532-1866022657.1577780532 refrence https://docs.opencv.org/4.1.2/d8/d23/classcv_1_1Moments.html#a8b1b4917d1123abc3a3c16b007a7319b https://github.com/opencv/opencv/blob/b6a58818bb6b30a1f9d982b3f3f53228ea5a13c1/modules/imgproc/src/moments.cpp''' huMomentLists=[] for img in imgLists: m_=cv2.moments(img) hu = cv2.HuMoments(m_) huMomentLists.append(np.array(hu)) # normal hu matrix huMomentLists = np.abs(huMomentLists) huMomentLists = np.log(huMomentLists) huMomentLists = np.abs(huMomentLists) return imgLists,textLists,huMomentLists if __name__=='__main__': img=cv2.imread(r'.\img\describe.png') # img=draw_contours(img) # cv2.imshow("contours",img) imgReturn,contours=get_region_description(img) i=0 for cnt in contours: i+=1 print("第%d个区域信息："%i) print("质心：",cnt[0]) print("方向：%f度"%cnt[1]) print("面积：", cnt[2]) print("周长：", cnt[3]) print("圆形度：", cnt[4]) print("矩形度：", cnt[5]) cv2.imshow("result",imgReturn) cv2.waitKey(0)