python - 如何用python对扫描页面进行逐字切片？

Question

有什么方法可以将文本的扫描图像分割成多个图像，每个图像包含一个单词？即，如果我们扫描包含“n”个词的页面，那么脚本应该生成“n”个单独的图像。

(使用 python )

Answer 1

这不是我非常熟悉的领域，但假设您无法使用 OCR (因为您的文字难以辨认或其他原因)，我会(可能天真地)尝试类似的方法:

• 将图像数据加载到内存中
• 将像素数据拆分为图像的行
• 找到所有只有白色像素的“行”:将这些标记为“白色行”
• 为每个“白行”中的每个“列”尝试找到白色间隙
• 采用所有新的 x,y 坐标并剪切图像。

实际上，这听起来像是一个有趣的练习，所以我试了一下 pyPNG模块:

import png
import sys

KERNING = 3

def find_rows(pixels,width, height):
    "find all rows that are purely white"
    white_rows = []
    is_white = False
    for y in range(height):
        if sum(sum( pixels[(y*4*width)+x*4+p] for p in range(3)) for x in range(width)) >= width*3*254:
            if not is_white:
                white_rows.append(y)
            is_white = True
        else:
            is_white = False
    return white_rows

def find_words_in_image(blob, tolerance=30):    
    n = 0
    r = png.Reader(bytes=blob)
    (width,height,pixels_rows,meta) = r.asRGBA8()
    pixels = []
    for row in pixels_rows:
        for px in row:
            pixels.append(px)
    # find each horizontal line
    white_rows = find_rows(pixels,width,height)
    # for each line try to find a white vertical gap
    for i,y in enumerate(white_rows):
        if y >= len(white_rows):
            continue
        y2 = white_rows[i+1]
        height_of_row = y2 - y
        is_white = False
        white_cols = []
        last_black = -100
        for x in range(width-4):
            s = y*4*width+x*4
            if sum(pixels[s+y3*4*width] + pixels[s+y3*4*width+1] + pixels[s+y3*4*width+2] for y3 in range(height_of_row)) >= height_of_row*3*240:
                if not is_white:
                    if len(white_cols)>0 and x-last_black < KERNING:
                        continue
                    white_cols.append(x)
                is_white = True
            else:
                is_white = False
                last_black = x
        # now we have a list of x,y co-oords for all the words on this row
        for j,x in enumerate(white_cols):
            if j >= len(white_cols)-1:
                continue
            wordpx = []
            new_width = white_cols[j+1]-x
            new_height = y2-y
            x_offset = x*4
            for h in range(new_height):
                y_offset = (y+h)*4*width
                start = x_offset+y_offset
                wordpx.append( pixels[start:start+(new_width*4)] )
            n += 1
            with open('word%s.png' % n, 'w') as f:
                w = png.Writer(
                    width=new_width,
                    height=new_height,
                    alpha=True
                    )
                w.write(f,wordpx)
    return n



if __name__ == "__main__":
    #
    # USAGE: python png2words.py yourpic.png
    #
    # OUTPUT: [word1.png...word2.png...wordN.png]
    #
    n = find_words_in_image( open(sys.argv[1]).read() )
    print "found %s words" % n