def tight_word(gray, rect):
def _get_row(timg):
v, groups, widths = preprocess.line_shadow(timg)
index = np.argmax(widths)
return groups[index]
def _get_col(timg):
v, groups, widths = preprocess.column_shadow(timg)
index = np.argmax(widths)
return groups[index]
x, y, w, h = rect
img = gray[y:y + h, x:x + w]
threshold = utils.custom_threshold(img)
# if w > h/1.8: # 先垂直再水平
# x0, x1 = _get_col(threshold)
# y0, y1 = _get_row(threshold[0:h, x0:x1])
# else: # 先水平再垂直
y0, y1 = _get_row(threshold)
x0, x1 = _get_col(threshold[y0:y1 + 1, 0:w])
y2, y3 = _get_row(threshold[y0:y1 + 1, x0:x1 + 1])
cv2.rectangle(threshold, (x0, y0 + y2), (x1 + 1, y0 + y2 + y3 + 1),
(255, 255, 255), 2) # 用矩形显示最终字符
cv2.imshow('tight{}'.format(x), threshold)
return (x0 + x, y0 + y2 + y, x1 - x0 + 1, y3 - y2 + 1)
def rect_boundary(grayImg):
# 数字区域定位
thresh = utils.custom_threshold(grayImg) # 二值化
# 取最中间的一段图像做水平投影，取出黑色占比最大的一段
height, width = grayImg.shape[:2]
middleImg = grayImg[0:height, (width/2-50):(width/2+50)]
cv2.imshow('middle', middleImg)
def pre_process(img, show=False):
img = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY) # 把输入图像灰度化
img = utils.custom_threshold(img)
if show:
cv2.imshow("thresh", img)
# kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (4, 4)) # 形态学处理:定义矩形结构
# img = cv2.erode(img, kernel, iterations=2) # 腐蚀
# if show:
# cv2.imshow("erode", img)
#
# kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (4, 4)) # 形态学处理:定义矩形结构
# img = cv2.dilate(img, kernel, iterations=2)
# if show:
# cv2.imshow("dilate", img)
return img
def pre_process(img, show=False):
# img = preprocess.convert_red_to_black(img)
img = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY) # 把输入图像灰度化
gris = cv2.GaussianBlur(img, (3, 3), 0) # 高斯滤波
img = utils.custom_threshold(gris)
if show:
cv2.imshow("thresh", img)
# kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (4, 4)) # 形态学处理:定义矩形结构
# img = cv2.erode(img, kernel, iterations=2) # 腐蚀
# if show:
# cv2.imshow("erode", img)
#
# kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (4, 4)) # 形态学处理:定义矩形结构
# img = cv2.dilate(img, kernel, iterations=2)
# if show:
# cv2.imshow("dilate", img)
return img
import cv2
import numpy as np
import utils
img = cv2.imread('./test0309/36.jpg')
gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
cv2.imshow('gray', gray)
edge_output = cv2.Canny(gray, 50, 150)
cv2.imshow("Canny Edge", edge_output)
blur = cv2.blur(gray, (3, 3))
cv2.imshow('blur', blur)
gauss = cv2.GaussianBlur(gray, (3, 3), 0)
cv2.imshow('guass', gauss)
kernel = np.ones((5, 5), np.float32) / 25
# gray = cv2.filter2D(gray, -1, kernel)
gray = cv2.medianBlur(gray, 5)
threshold = utils.custom_threshold(gray)
cv2.imshow('threshold', threshold)
cv2.imshow('filter2D', gray)
cv2.waitKey(0)
def img_to_words(img, show=False, words=6):
# # 将图片分割成字符
# 步骤1：识别出数字区域
(x1, y1, w1, h1) = _rect_digital(img)
digital1 = img[y1:y1 + h1, x1:x1 + w1]
if show:
cv2.imshow('first digital area', digital1)
# 步骤1：取出数字区域
(x2, y2, w2, h2) = _rect_digital(digital1)
x = x1 + x2
y = y1 + y2
w = w2
h = h2
# 步骤2：图像预处理
# closed = pre_process(img, show)
gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY) # 把输入图像灰度化
gray = cv2.GaussianBlur(gray, (3, 3), 0) # 高斯滤波
closed = utils.custom_threshold(gray)
closed = closed[y:y + h, x:x + w]
# 步骤3：进行列分割，分割的图像基于数字区
xWords = column_slit2(closed, words=words, show=show)
# xWords = column_split(closed, show) # 垂直的字符位置
cv2.imshow('closed', closed)
if len(xWords) == 0:
print 'column_split return empty'
return []
tmp = [] # 最终的字符的(x,y,w,h)
# 步骤4：将列分割的结果进行行分割，分解出最终的数字
wordHeights = []
for i in range(0, len(xWords)):
xPos = xWords[i]
# 对每个字符进行水平分割
wordRect = closed[0:h, xPos[0]:xPos[1] + 1]
# wordRect = utils.custom_threshold(wordRect)
yPos = line_split(wordRect, show_window=None)
if yPos is None:
continue
# 对字符再做一次垂直头像，取出左右不必要的空间
r1 = wordRect[yPos[0]:yPos[1] + 1, 0:wordRect.shape[1]]
xPos2 = word_column_split(r1)
if xPos2 is None:
continue
# x1 = xPos[0] + x
# y1 = yPos[0] + y
# w1 = xPos[1] - xPos[0]
# h1 = yPos[1] - yPos[0]
x1 = xPos[0] + xPos2[0] + x
y1 = yPos[0] + y
w1 = xPos2[1] - xPos2[0] + 1
h1 = yPos[1] - yPos[0] + 1
tmp.append((x1, y1, w1, h1))
wordHeights.append(h1)
# 步骤5：对切割结果过滤，去除不可能是数字的部分：取字符高度的中位数，去除高度小于高度中位数1/3的字符
heightMedium = np.average(wordHeights)
wordRects = []
for t in tmp:
(x1, y1, w1, h1) = t
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)
resizedHeight = int(oriHeight / (oriWidth / float(800)))
# 2、大小归一化，宽度固定为800
img = cv2.resize(img, (800, resizedHeight)) # 将图片宽度固定为800
# 3、字符分割
wordRects = wordSplit.img_to_words(img, show) # 字符分割
# 4、图像灰化后颜色反转
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 = []
index = 1
for (x, y, w, h) in wordRects:
if w == 0 or h == 0:
continue
roi = gray[y:y + h, x:x + w]
roi = utils.custom_threshold(roi)
cv2.imshow('roi{}'.format(index), roi)
roi = cv2.resize(roi, (width, height))
index += 1
roi_small = roi.reshape((1, width * height))
roi_small = np.float32(roi_small)
retval, results, neigh_resp, dists = model.findNearest(roi_small,
k=1)
responseNmber = int((results[0][0]))
if responseNmber > 9:
print u'识别到刻度'
# continue
cv2.rectangle(img, (x, y), (x + w, y + h), (0, 0, 255), 2)
cv2.putText(img, str(responseNmber), (x + 10, y + 25),
cv2.FONT_HERSHEY_SIMPLEX, 1.2, (0, 255, 0), 2)
if show: