python+opencv-tessert OCR 实现简易的车牌的检测与识别

# 车牌识别 import cv2 as cv import numpy as np import pytesseract as tess from PIL import Image def license_prepation(image): image_hsv = cv.cvtColor(image, cv.COLOR_BGR2HSV) # 将图像转换为hsv色彩空间 low_hsv = np.array([108, 43, 46]) # 设置颜色 high_hsv = np.array([124, 255, 255]) mask = cv.inRange(image_hsv, lowerb=low_hsv, upperb=high_hsv) # cv.inrange 函数用来追踪image image_dst = cv.bitwise_and(image, image, mask=mask) # 取frame与mask中不为0的相与，mask=mask必须有 # cv.imshow('license_dst', image_dst) image_blur = cv.GaussianBlur(image_dst, (7, 7), 0) # cv.imshow('license_blur',image_blur) image_gray = cv.cvtColor(image_blur, cv.COLOR_BGR2GRAY) ret, binary = cv.threshold(image_gray, 0, 255, cv.THRESH_BINARY | cv.THRESH_OTSU) # cv.imshow('binary', binary) kernel1 = cv.getStructuringElement(cv.MORPH_RECT, (4, 6)) image_opened = cv.morphologyEx(binary, cv.MORPH_OPEN, kernel1) # cv.imshow('license_opened', image_opened) kernel2 = cv.getStructuringElement(cv.MORPH_RECT, (7, 7)) image_closed = cv.morphologyEx(image_opened, cv.MORPH_CLOSE, kernel2) # cv.imshow('license_closed', image_closed) return image_closed def choose_license_area(contours, Min_Area): temp_contours = [] for contour in contours: if cv.contourArea(contour) > Min_Area: # 面积大于MIN_AREA的区域保留 temp_contours.append(contour) license_area = [] for temp_contour in temp_contours: rect_tupple = cv.minAreaRect(temp_contour) # print(rect_tupple) rect_width, rect_height = rect_tupple[1] # 0为中心点，1为长和宽，2为角度 if rect_width < rect_height: rect_width, rect_height = rect_height, rect_width aspect_ratio = rect_width / rect_height # 车牌正常情况下宽高比在2 - 5.5之间 if aspect_ratio > 2 and aspect_ratio < 5.5: license_area.append(temp_contour) rect_vertices = cv.boxPoints(rect_tupple) rect_vertices = np.int0(rect_vertices) return license_area def license_segment(license_area): if (len(license_area)) == 1: for car_plate in license_area: # print(car_plate.shape) # print(car_plate) row_min, col_min = np.min(car_plate[:, 0, :], axis=0) # 行是row 列是col row_max, col_max = np.max(car_plate[:, 0, :], axis=0) # cv.rectangle(license, (row_min, col_min), (row_max, col_max), (0, 255, 0),2) card_img = license[col_min:col_max, row_min:row_max, :] # cv.imshow("card_img", card_img) cv.imwrite("card_img.jpg", card_img) return card_img def recognize_text(image): gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY) ret, binary = cv.threshold(gray, 120, 255, cv.THRESH_BINARY_INV) cv.imshow('bin', binary) # 白底黑字 bin1 = cv.resize(binary, (370, 82)) kernel1 = cv.getStructuringElement(cv.MORPH_RECT, (2, 5)) dilated = cv.dilate(bin1, kernel1) # 白色膨胀 text = tess.image_to_string(dilated, lang='chi_sim') print('识别的结果为:%s' % text) src = cv.imread("D:/opencv/image/license.png") # 打开要识别的照片，不能有中文路径 # print(src.shape) license = cv.resize(src, (800, int(800 * src.shape[0] / src.shape[1]))) # print(license.shape) cv.namedWindow('inputImage', 0) cv.imshow('inputImage', license) license_prepared = license_prepation(license) # cv.imshow('license_prepared', license_prepared) contours, hierarchy = cv.findContours(license_prepared, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE) Min_Area = 2000 license_area = choose_license_area(contours, Min_Area) result = license_segment(license_area) cv.imshow('result', result) recognize_text(result) cv.waitKey(0) cv.destroyAllWindows()