How to perform word break on a scanned page using python?

Question

Is it possible to cut a scanned image of text into several images containing one word each? those. if we crawl the page using "n" words, then the script should create "n" separate images.

(using python)

Answer 1

This is not an area in which I am very familer, but assuming you cannot use OCR (because your text is illegible or something), I would (probably naively) try something like:

• load image data into memory
• splitting pixel data into image lines
• find each “line” that has only white pixels: notice them as “white lines”
• For each column in each white row, try to find white spaces.
• take all your new x, y coordinates and crop the image.

It actually sounded like a fun exercise, so I gave it a project with pyPNG :

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 

Answer 2

You need to look at Blob Detection , this is an image processing technology. Also, this question has nothing to do with python, but looking for python blob detection libraries may help.