图像ocr识别（一）

研究了点OCR识别，本文讲下opencv方式-找出字符区域,虽然还不完善，但是记录下，后续往CNN+RNN+CTC方向走，此处就作为练手了。

效果1：

效果2：

效果3：

效果4（识别率不太好，只把大框识别了，字符的分割有问题）：

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

import cv2
import imutils
import numpy as np
from imageio import imread
import math
import matplotlib.pyplot as plt
 
 
def point_distance(p1, p2):
    return math.sqrt(math.pow(p2[0] - p1[0], 2) + math.pow(p2[1] - p1[1], 2))
 
 
def calc_height_width(box):
    width = point_distance(box[1], box[0])
    height = point_distance(box[0], box[3])
    return (width, height)
 
 
fileName = 'test1'
 
img = imread('imgs\\' + fileName + '.jpg')
img = imutils.resize(img, width=1920, height=2080)
 
 
 
cannyImg = cv2.Canny(img, 200, 200)
 
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
closed = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)
eroded = cv2.erode(closed, kernel)
 
cannyImg = cv2.Canny(eroded, 200, 200)
blurred = cv2.GaussianBlur(cannyImg, (105, 105), 0)
# blurred = cv2.GaussianBlur(cannyImg, (15, 15),0)
 
 
_, skin = cv2.threshold(blurred, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
 
 
contours, hierarchy = cv2.findContours(skin, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
 
contours = sorted(contours, key=cv2.contourArea, reverse=True)
 
boxes = []
for i in range(len(contours)):
    c = contours[i]
    rect = cv2.minAreaRect(c)
    box = np.int0(cv2.boxPoints(rect))
    (w, h) = calc_height_width(box)
    if w == 0 or h == 0:
        continue
    if w < 20 or h < 20:
        continue
    # boxes.append(box)
    rate1 = h / w * 100
    rate2 = w / h * 100
    if (10 <= rate1 <= 20) or (10 <= rate2 <= 20):
        print((w, h), '--------', rate1, '%', rate2, '%')
        boxes.append(box)
 
img = img.copy()
i = 0
 
 
def parse_chars(positions, min_thresh, min_range, max_range):
    charInfos = []
    begin = 0
    end = 0
    for idx in range(len(positions)):
        if positions[idx] > min_thresh and begin == 0:
            begin = idx
        elif positions[idx] > min_thresh and begin != 0:
            if idx - begin > max_range:
                charInfo = {'begin': begin, 'end': idx}
                charInfos.append(charInfo)
 
                begin = 0
                end = 0
            continue
        elif positions[idx] < min_thresh and begin != 0:
            end = idx
            if end - begin >= min_range:
                charInfo = {'begin': begin, 'end': end}
                charInfos.append(charInfo)
 
                begin = 0
                end = 0
        elif positions[idx] < min_thresh or begin == 0:
            continue
 
    return charInfos
 
 
def process_more(windowName, imgSrc):
    ori_imgSrc = imgSrc.copy()
    # cv2.imshow(windowName+'111', ori_imgSrc)
 
    kernel = cv2.getStructuringElement(cv2.MORPH_CROSS, (3, 3))
    closed = cv2.morphologyEx(imgSrc, cv2.MORPH_CLOSE, kernel)
    imgSrc = cv2.erode(closed, kernel)
 
    imgSrc = cv2.Canny(imgSrc, 300, 300)
    kernel = np.ones((5, 5), np.uint8)
    imgSrc = cv2.dilate(imgSrc, kernel, iterations=1)
    _, imgSrc = cv2.threshold(imgSrc, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
 
    # rows---> ori_imgSrc.shape[0]
    # cols---> ori_imgSrc.shape[1]
    rows = ori_imgSrc.shape[0]
    cols = ori_imgSrc.shape[1]
    tags = np.zeros((cols))
 
    for row in range(rows):
        for col in range(cols):
            if imgSrc[row][col] == 255:
                tags[col] += 1
 
    char_positions = parse_chars(positions=tags, min_thresh=8, min_range=25, max_range=100)
<br>    print(len(char_positions))
    for p in char_positions:
        leftTop = (p['begin'], 0)
        rightBottom = (p['end'], rows - 2)
        cv2.rectangle(ori_imgSrc, (leftTop[0], leftTop[1]), (rightBottom[0], rightBottom[1]), (0, 255, 0), 2)
 
    ori_imgSrc = imutils.resize(ori_imgSrc, width=450)
    cv2.imshow(windowName, ori_imgSrc)
 
 
for box in boxes:
    # img = cv2.drawContours(img, [box], -1, (0, 0, 255), 3)
    x_from = np.min(box[:, 1])
    x_end = np.max(box[:, 1])
    y_from = np.min(box[:, 0])
    y_end = np.max(box[:, 0])
 
    if x_from < 0:
        x_from = 0
    if y_from < 0:
        y_from = 0
 
    img_tmp = img[x_from:x_end, y_from:y_end]
    # cv2.imshow("ffff111" + str(i), img_tmp)
 
    (w, h) = calc_height_width(box)
 
    if w > h:
        # 左上角, 左下角,右上角
        # 3,2,4
        matSrc = np.float32([
            [box[2][0], box[2][1]],
            [box[1][0], box[1][1]],
            [box[3][0], box[3][1]]
        ])
        matDst = np.float32([
            [0, 0],
            [0, h],
            [w, 0]
        ])
        matAffine = cv2.getAffineTransform(matSrc, matDst)
        dst = cv2.warpAffine(img, matAffine, (int(w), int(h)))
    else:
        # 左上角, 左下角,右上角
        # 右上角, 左上角, 右下角
        # 3,2,4
        # 4,3,1
        matSrc = np.float32([
            [box[3][0], box[3][1]],
            [box[2][0], box[2][1]],
            [box[0][0], box[0][1]]
        ])
        matDst = np.float32([
            [0, 0],
            [0, w],
            [h, 0]
        ])
        matAffine = cv2.getAffineTransform(matSrc, matDst)
        dst = cv2.warpAffine(img, matAffine, (int(h), int(w)))
 
    process_more("ffff222asdfas" + str(i), dst.copy())
 
    i += 1
 
img = imutils.resize(img, width=600, height=600)
cv2.imshow("Frame6", img)
 
cv2.waitKey(100000) & 0xFF
 
cv2.destroyAllWindows()