def rect_boundary2(img, show=False): # 通过HSV获取黑色部分
converted = preprocess.convert_red_to_black(img, show)
mat = preprocess.get_black_mat(converted, show=show)
angle, rotated = utils.correct_skew(mat, is_gray=True)
if show:
cv2.imshow('rotated', rotated)
# 水平投影
v, groups, counts = preprocess.line_shadow(rotated, color=255)
if h1 >= heightMedium / 2:
wordRects.append(t)
# 步骤6：通过字符的宽度判断是否包含多个字符
wordRects = _validate_by_width(closed, wordRects, words=words)
# 步骤7：通过间隔判断不合理的数据
wordRects = _validate_by_interval(wordRects)
# 步骤8：重新二值化后再次收缩字符范围
for i in range(0, len(wordRects)):
r = wordRects[i]
r1 = tight_word(gray, r)
wordRects[i] = r1
if show:
for (x1, y1, w1, h1) in wordRects:
cv2.rectangle(img, (x1, y1), (x1 + w1, y1 + h1), (0, 0, 255),
2) # 用矩形显示最终字符
cv2.imshow('words', img)
return wordRects # 返回每个字符的(x,y,w,h)
if __name__ == '__main__':
# img = cv2.imread('./test0309/34.jpg') # 4,29,34,36,37,44,45,46
# img = cv2.imread('./area/13.png') # 5，18，31，42，51，54，55
img = cv2.imread('./area/1.png') # 34,37,45,46
angle, img = utils.correct_skew(img, is_gray=False)
cv2.imshow('skew', img)
oriHeight, oriWidth = img.shape[:2]
resizedHeight = int(oriHeight / (oriWidth / float(800)))
img = cv2.resize(img, (800, resizedHeight)) # 将图片宽度固定为800
img_to_words(img, show=True, words=6)
cv2.waitKey(0)
cv2.destroyAllWindows()
import numpy as np
import wordSplit
import utils
#TRAINING---------------------------------------------------------------------------------------------------------------
samples = np.loadtxt('./data/general_samples.data', np.float32)
responses = np.loadtxt('./data/general_responses.data', np.float32)
responses = responses.reshape((responses.size, 1))
model = cv2.ml.KNearest_create()
model.train(samples, cv2.ml.ROW_SAMPLE, responses)
if __name__ == '__main__':
show = False
imgName = '26.png'
img = cv2.imread('./area/{}'.format(imgName))
angle, img = utils.correct_skew(img)
oriHeight, oriWidth = img.shape[:2]
resizedHeight = int(oriHeight / (oriWidth / float(800)))
img = cv2.resize(img, (800, resizedHeight)) # 将图片宽度固定为800
wordRects = wordSplit.img_to_words(img, show) # 字符分割
gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY) # 把输入图像灰度化
utils.color_reverse(gray)
gris = cv2.GaussianBlur(gray, (3, 3), 0) # 高斯滤波
chars = []
for (x, y, w, h) in wordRects:
roi = gris[y:y + h, x:x + w]
roi = support_library.recon_borde(roi)
roi_small = cv2.resize(roi, (10, 10))
roi_small = roi_small.reshape((1, 100))
roi_small = np.float32(roi_small)
easyreader = easyocr.Reader(['en'])
easyresult = easyreader.readtext("cropped_image.jpg")
easywiskew = []
for e in easyresult:
easywiskew.append(e[1].upper())
pipeline = keras_ocr.pipeline.Pipeline()
images = ['cropped_image.jpg']
prediction = pipeline.recognize(images)[0]
keraswiskew = []
for p in prediction:
keraswiskew.append(p[0].upper())
# Correcting Skewness
image = cv2.imread("cropped_image.jpg")
angle, rotated = utils.correct_skew(image)
cv2.imwrite("rotated_image.jpg", rotated)
# With Skewness outputs
easyreader = easyocr.Reader(['en'])
easyresult = easyreader.readtext("rotated_image.jpg")
easynoskew = []
for e in easyresult:
easynoskew.append(e[1].upper())
pipeline = keras_ocr.pipeline.Pipeline()
images = ["rotated_image.jpg"]
prediction = pipeline.recognize(images)[0]
kerasnoskew = []
for p in prediction:
kerasnoskew.append(p[0].upper())