基于投影的图像文字分割 - python

环境

python 3.6
python-opencv 3.4.0

思路

对图片进行水平方向投影，得到每一行文字的起止线
对每一行文字进行竖直方向投影，若有连续的像素满足条件，则为一个字，进行分割

代码

import cv2
import numpy as np

path = 'd:\\testdata\\test\\bin1.png'
root = 'd:\\testdata\\test\\'
dsize = 28 #归一化处理的图像大小
img = cv2.imread(path)
data = np.array(img)
len_x = data.shape[0] 
len_y = data.shape[1]
min_val = 10 #设置最小的文字像素高度，防止切分噪音字符

start_i = -1
end_i = -1
rowPairs = [] #存放每行的起止坐标

#行分割
for i in range(len_x):
    if(not data[i].all() and start_i < 0): 
        start_i = i
    elif(not data[i].all()):
        end_i = i
    elif (data[i].all() and start_i >= 0):
        #print(end_i - start_i)
        if(end_i - start_i >= min_val):
            rowPairs.append((start_i, end_i))
        start_i, end_i = -1, -1

#print(rowPairs)

#列分割
start_j = -1
end_j = -1
min_val_word = 5  #最小文字像素长度
number = 0 #分割后保存编号
for start, end in rowPairs:
    for j in range(len_y):
        if(not data[start: end, j].all() and start_j < 0):
            start_j = j
        elif(not data[start: end, j].all()):
             end_j = j
        elif(data[start: end, j].all() and start_j >= 0):
            if(end_j - start_j >= min_val_word):
                #print(end_j - start_j)
                tmp = data[start:end, start_j: end_j]
                im2save = cv2.resize(tmp, (dsize,dsize)) #归一化处理
                cv2.imwrite(root + '%d.png' % number, im2save) 
                number += 1
                #print("%d  pic" % number)
            start_j, end_j = -1, -1

结果

分割前：

分割后：

总揽：

单例：

效果还行