首先這是一個成語填字遊戲，大概就是一張成語填字遊戲圖片，通過opencv圖像識別後轉爲矩陣，再通過解算法，解出答案，在顯示到圖片上。
源代碼：https://github.com/mayue801/crossword-puzzle--idiom
本文采用投影分割法對印刷體漢字進行分割。
投影分割是先水平方向投影，在豎直方向投影，或者先豎直方向再水平方向投影。本文選用先豎直，再水平。
1.豎直投影。
------------
----------------
代碼：
#針對的是印刷版的漢字，所以採用了投影法分割
#此函數是行分割，結果是一行文字
def YShadow(path):
img = cv2.imread(path) #原圖像
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY) #灰度圖像
height,width = img.shape[:2]
#blur = cv2.GaussianBlur(gray,(5,5),0) #高斯模糊
blur = cv2.blur(gray,(8,8)) #均值模糊
thresh = cv2.adaptiveThreshold(blur,255,1,1,11,2) #自適應閾值分割
temp = thresh
if(width > 500 and height > 400): #圖像字體較小時，需要進行膨脹操作
kernel = np.ones((5,5),np.uint8) #卷積核
dilation = cv2.dilate(thresh,kernel,iterations = 1) #膨脹操作使得單個文字圖像被黑像素填充
temp = dilation
'''
cv2.imshow('image',temp)
cv2.waitKey(0)
cv2.destroyAllWindows()
'''
perPixelValue = 1 #每個像素的值
projectValArry = np.zeros(width, np.int8) #創建一個用於儲存每列黑色像素個數的數組
for i in range(0,height):
for j in range(0,width):
perPixelValue = temp[i,j]
if (perPixelValue == 255): #如果是黑字，對應位置的值+1
projectValArry[i] += 1
# print(projectValArry[i])
canvas = np.zeros((height,width), dtype="uint8")
for i in range(0,height):
for j in range(0,width):
perPixelValue = 255 #白色背景
canvas[i, j] = perPixelValue
for i in range(0,height):
for j in range(0,projectValArry[i]):
perPixelValue = 0 #黑色直方圖投影
canvas[i, width-j-1] = perPixelValue
'''
cv2.imshow('canvas',canvas)
cv2.waitKey(0)
cv2.destroyAllWindows()
'''
list = []
startIndex = 0 #記錄進入字符區的索引
endIndex = 0 #記錄進入空白區域的索引
inBlock = 0 #是否遍歷到了字符區內
for i in range(height):
if (inBlock == 0 and projectValArry[i] != 0): #進入字符區
inBlock = 1
startIndex = i
elif (inBlock == 1 and projectValArry[i] == 0):#進入空白區
endIndex = i
inBlock = 0
subImg = gray[startIndex:endIndex+1,0:width] #將對應字的圖片截取下來
#print(startIndex,endIndex+1)
list.append(subImg)#添加這個字圖像到list
#print(len(list))
return list 
#對行字進行單個字的分割
def XShadow(path):
img = cv2.imread(path)
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
height,width = img.shape[:2]
# print(height,width)
#blur = cv2.GaussianBlur(gray,(5,5),0)
blur = cv2.blur(gray,(8,8))
thresh = cv2.adaptiveThreshold(blur,255,1,1,11,2)
if(width > 500):
kernel = np.ones((4, 4),np.uint8) #卷積核
else:
kernel = np.ones((2, 2),np.uint8) #卷積核
dilation = cv2.dilate(thresh,kernel,iterations = 1) #膨脹操作使得單個文字圖像被黑像素填充
'''
cv2.imshow('image',thresh)
cv2.waitKey(0)
cv2.destroyAllWindows()
'''
perPixelValue = 1 #每個像素的值
projectValArry = np.zeros(width, np.int8) #創建一個用於儲存每列黑色像素個數的數組
for i in range(0,width):
for j in range(0,height):
perPixelValue = dilation[j,i]
if (perPixelValue == 255): #如果是黑字
projectValArry[i] += 1
# print(projectValArry[i])
canvas = np.zeros((height,width), dtype="uint8")
for i in range(0,width):
for j in range(0,height):
perPixelValue = 255 #白色背景
canvas[j, i] = perPixelValue
for i in range(0,width):
for j in range(0,projectValArry[i]):
perPixelValue = 0 #黑色直方圖投影
canvas[height-j-1, i] = perPixelValue
'''
cv2.imshow('canvas',canvas)
cv2.waitKey(0)
cv2.destroyAllWindows()
'''
list = []
startIndex = 0 #記錄進入字符區的索引
endIndex = 0 #記錄進入空白區域的索引
inBlock = 0 #是否遍歷到了字符區內
for i in range(width):
if (inBlock == 0 and projectValArry[i] != 0): #進入字符區
inBlock = 1
startIndex = i
elif (inBlock == 1 and projectValArry[i] == 0): #進入投影區
endIndex = i
inBlock = 0
#subImg = gray[0:height, startIndex:endIndex+1] #endIndex+1
#print(startIndex,endIndex+1)
list.append([startIndex, 0, endIndex-startIndex-1, height])
#print(len(list))
return list分割完後，將對應圖片樣本存儲到對應文件夾，每個字共10種樣本
將這些樣本及標記保存後，分別加載到samples.npy, label.npy中。供後續的機器學習算法訓練使用。
下篇講解填字圖片漢字的提取與機器學習算法訓練樣本，識別漢字的過程。