一、
灰度处理,就是把彩色的验证码图片转为灰色的图片。
二值化,是将图片处理为只有黑白两色的图片,利于后面的图像处理和识别
1 # 自适应阀值二值化
2 def _get_dynamic_binary_image(filedir, img_name):
3 filename = './out_img/' + img_name.split('.')[0] + '-binary.jpg'
4 img_name = filedir + '/' + img_name
5 print('.....' + img_name)
6 im =dz.imread(img_name)
7 im = dz.cvtColor(im,dz.COLOR_BGR2GRAY) #灰值化
8 # 二值化
9 th1 = dz.adaptiveThreshold(im, 255, dz.ADAPTIVE_THRESH_GAUSSIAN_C, dz.THRESH_BINARY, 21, 1)
10
11 dz.imwrite(filename,th1)
12 return th1
二、去除边框
1 # 去除边框
2 def clear_border(img,img_name):
3 filename = './out_img/' + img_name.split('.')[0] + '-clearBorder.jpg'
4 h, w = img.shape[:2]
5 for y in range(0, w):
6 for x in range(0, h):
7 if y < 2 or y > w - 2:
8 img[x, y] = 255
9 if x < 2 or x > h -2:
10 img[x, y] = 255
11
12 cv2.imwrite(filename,img)
13 return img
在用OpenCV时,图片的矩阵点是反的,就是长和宽是颠倒的
三、降噪
降噪是验证码处理中比较重要的一个步骤,我这里使用了点降噪和线降噪,,,只能去除细的干扰线
1 # 干扰线降噪
2 def interference_line(img, img_name):
3 filename = './out_img/' + img_name.split('.')[0] + '-interferenceline.jpg'
4 h, w = img.shape[:2]
5 # !!opencv矩阵点是反的
6 # img[1,2] 1:图片的高度,2:图片的宽度
7 for y in range(1, w - 1):
8 for x in range(1, h - 1):
9 count = 0
10 if img[x, y - 1] > 245:
11 count = count + 1
12 if img[x, y + 1] > 245:
13 count = count + 1
14 if img[x - 1, y] > 245:
15 count = count + 1
16 if img[x + 1, y] > 245:
17 count = count + 1
18 if count > 2:
19 img[x, y] = 255
20 cv2.imwrite(filename,img)
21 return img
1 # 点降噪
2 def interference_point(img,img_name, x = 0, y = 0):
3 """
4 9邻域框,以当前点为中心的田字框,黑点个数
5 :param x:
6 :param y:
7 :return:
8 """
9 filename = './out_img/' + img_name.split('.')[0] + '-interferencePoint.jpg'
10 # todo 判断图片的长宽度下限
11 cur_pixel = img[x,y]# 当前像素点的值
12 height,width = img.shape[:2]
13
14 for y in range(0, width - 1):
15 for x in range(0, height - 1):
16 if y == 0: # 第一行
17 if x == 0: # 左上顶点,4邻域
18 # 中心点旁边3个点
19 sum = int(cur_pixel) \
20 + int(img[x, y + 1]) \
21 + int(img[x + 1, y]) \
22 + int(img[x + 1, y + 1])
23 if sum <= 2 * 245:
24 img[x, y] = 0
25 elif x == height - 1: # 右上顶点
26 sum = int(cur_pixel) \
27 + int(img[x, y + 1]) \
28 + int(img[x - 1, y]) \
29 + int(img[x - 1, y + 1])
30 if sum <= 2 * 245:
31 img[x, y] = 0
32 else: # 最上非顶点,6邻域
33 sum = int(img[x - 1, y]) \
34 + int(img[x - 1, y + 1]) \
35 + int(cur_pixel) \
36 + int(img[x, y + 1]) \
37 + int(img[x + 1, y]) \
38 + int(img[x + 1, y + 1])
39 if sum <= 3 * 245:
40 img[x, y] = 0
41 elif y == width - 1: # 最下面一行
42 if x == 0: # 左下顶点
43 # 中心点旁边3个点
44 sum = int(cur_pixel) \
45 + int(img[x + 1, y]) \
46 + int(img[x + 1, y - 1]) \
47 + int(img[x, y - 1])
48 if sum <= 2 * 245:
49 img[x, y] = 0
50 elif x == height - 1: # 右下顶点
51 sum = int(cur_pixel) \
52 + int(img[x, y - 1]) \
53 + int(img[x - 1, y]) \
54 + int(img[x - 1, y - 1])
55
56 if sum <= 2 * 245:
57 img[x, y] = 0
58 else: # 最下非顶点,6邻域
59 sum = int(cur_pixel) \
60 + int(img[x - 1, y]) \
61 + int(img[x + 1, y]) \
62 + int(img[x, y - 1]) \
63 + int(img[x - 1, y - 1]) \
64 + int(img[x + 1, y - 1])
65 if sum <= 3 * 245:
66 img[x, y] = 0
67 else: # y不在边界
68 if x == 0: # 左边非顶点
69 sum = int(img[x, y - 1]) \
70 + int(cur_pixel) \
71 + int(img[x, y + 1]) \
72 + int(img[x + 1, y - 1]) \
73 + int(img[x + 1, y]) \
74 + int(img[x + 1, y + 1])
75
76 if sum <= 3 * 245:
77 img[x, y] = 0
78 elif x == height - 1: # 右边非顶点
79 sum = int(img[x, y - 1]) \
80 + int(cur_pixel) \
81 + int(img[x, y + 1]) \
82 + int(img[x - 1, y - 1]) \
83 + int(img[x - 1, y]) \
84 + int(img[x - 1, y + 1])
85
86 if sum <= 3 * 245:
87 img[x, y] = 0
88 else: # 具备9领域条件的
89 sum = int(img[x - 1, y - 1]) \
90 + int(img[x - 1, y]) \
91 + int(img[x - 1, y + 1]) \
92 + int(img[x, y - 1]) \
93 + int(cur_pixel) \
94 + int(img[x, y + 1]) \
95 + int(img[x + 1, y - 1]) \
96 + int(img[x + 1, y]) \
97 + int(img[x + 1, y + 1])
98 if sum <= 4 * 245:
99 img[x, y] = 0
100 cv2.imwrite(filename,img)
101 return img
五、字符切割
1 def cfs(im,x_fd,y_fd):
2 '''用队列和集合记录遍历过的像素坐标代替单纯递归以解决cfs访问过深问题
3 '''
4
5 # print('**********')
6
7 xaxis=[]
8 yaxis=[]
9 visited =set()
10 q = Queue()
11 q.put((x_fd, y_fd))
12 visited.add((x_fd, y_fd))
13 offsets=[(1, 0), (0, 1), (-1, 0), (0, -1)]#四邻域
14
15 while not q.empty():
16 x,y=q.get()
17
18 for xoffset,yoffset in offsets:
19 x_neighbor,y_neighbor = x+xoffset,y+yoffset
20
21 if (x_neighbor,y_neighbor) in (visited):
22 continue # 已经访问过了
23
24 visited.add((x_neighbor, y_neighbor))
25
26 try:
27 if im[x_neighbor, y_neighbor] == 0:
28 xaxis.append(x_neighbor)
29 yaxis.append(y_neighbor)
30 q.put((x_neighbor,y_neighbor))
31
32 except IndexError:
33 pass
34 # print(xaxis)
35 if (len(xaxis) == 0 | len(yaxis) == 0):
36 xmax = x_fd + 1
37 xmin = x_fd
38 ymax = y_fd + 1
39 ymin = y_fd
40
41 else:
42 xmax = max(xaxis)
43 xmin = min(xaxis)
44 ymax = max(yaxis)
45 ymin = min(yaxis)
46 #ymin,ymax=sort(yaxis)
47
48 return ymax,ymin,xmax,xmin
49
50 def detectFgPix(im,xmax):
51 '''搜索区块起点
52 '''
53
54 h,w = im.shape[:2]
55 for y_fd in range(xmax+1,w):
56 for x_fd in range(h):
57 if im[x_fd,y_fd] == 0:
58 return x_fd,y_fd
59
60 def CFS(im):
61 '''切割字符位置
62 '''
63
64 zoneL=[]#各区块长度L列表
65 zoneWB=[]#各区块的X轴[起始,终点]列表
66 zoneHB=[]#各区块的Y轴[起始,终点]列表
67
68 xmax=0#上一区块结束黑点横坐标,这里是初始化
69 for i in range(10):
70
71 try:
72 x_fd,y_fd = detectFgPix(im,xmax)
73 # print(y_fd,x_fd)
74 xmax,xmin,ymax,ymin=cfs(im,x_fd,y_fd)
75 L = xmax - xmin
76 H = ymax - ymin
77 zoneL.append(L)
78 zoneWB.append([xmin,xmax])
79 zoneHB.append([ymin,ymax])
80
81 except TypeError:
82 return zoneL,zoneWB,zoneHB
83
84 return zoneL,zoneWB,zoneHB
切割粘连字符代码
1 # 切割的位置
2 im_position = CFS(im)
3
4 maxL = max(im_position[0])
5 minL = min(im_position[0])
6
7 # 如果有粘连字符,如果一个字符的长度过长就认为是粘连字符,并从中间进行切割
8 if(maxL > minL + minL * 0.7):
9 maxL_index = im_position[0].index(maxL)
10 minL_index = im_position[0].index(minL)
11 # 设置字符的宽度
12 im_position[0][maxL_index] = maxL // 2
13 im_position[0].insert(maxL_index + 1, maxL // 2)
14 # 设置字符X轴[起始,终点]位置
15 im_position[1][maxL_index][1] = im_position[1][maxL_index][0] + maxL // 2
16 im_position[1].insert(maxL_index + 1, [im_position[1][maxL_index][1] + 1, im_position[1][maxL_index][1] + 1 + maxL // 2])
17 # 设置字符的Y轴[起始,终点]位置
18 im_position[2].insert(maxL_index + 1, im_position[2][maxL_index])
19
20 # 切割字符,要想切得好就得配置参数,通常 1 or 2 就可以
21 cutting_img(im,im_position,img_name,1,1
切割粘连字符代码
1 def cutting_img(im,im_position,img,xoffset = 1,yoffset = 1):
2 filename = './out_img/' + img.split('.')[0]
3 # 识别出的字符个数
4 im_number = len(im_position[1])
5 # 切割字符
6 for i in range(im_number):
7 im_start_X = im_position[1][i][0] - xoffset
8 im_end_X = im_position[1][i][1] + xoffset
9 im_start_Y = im_position[2][i][0] - yoffset
10 im_end_Y = im_position[2][i][1] + yoffset
11 cropped = im[im_start_Y:im_end_Y, im_start_X:im_end_X]
12 cv2.imwrite(filename + '-cutting-' + str(i) + '.jpg',cropped)
六、识别:
识别用的是typesseract库,主要识别一行字符和单个字符时的参数设置,识别中英文的参数设置,代码很简单就一行,我这里大多是filter文件的操作
1 # 识别验证码
2 cutting_img_num = 0
3 for file in os.listdir('./out_img'):
4 str_img = ''
5 if fnmatch(file, '%s-cutting-*.jpg' % img_name.split('.')[0]):
6 cutting_img_num += 1
7 for i in range(cutting_img_num):
8 try:
9 file = './out_img/%s-cutting-%s.jpg' % (img_name.split('.')[0], i)
10 # 识别字符
11 str_img = str_img + image_to_string(Image.open(file),lang = 'eng', config='-psm 10') #单个字符是10,一行文本是7
12 except Exception as err:
13 pass
14 print('切图:%s' % cutting_img_num)
15 print('识别为:%s' % str_img