Read the image of the puzzle game and output the sequence of each block

Purpose

For example, I want to read the image below and determine the arrangement of each block.

problem

I tried to get the value by Template Matching using OpenCV, but it takes time to match each image, and all the images with high correlation within the threshold value are output, so select them. The cost was high because I had to write a program.

policy

Using the RGB data in the PIL library, the RGB data of the image cut out for each block is correlated and recognized as the color closest to the reference value.

Implementation

Use the PIL Image object.

import numpy
from PIL import Image

def get_rgb(pic, box=""):
    if box == "":
        box = (0, 0, pic.width, pic.height)
    rgbimg = pic.crop(box).convert("RGB")
    rgb = np.array(rgbimg.getdata())
    return [__round(rgb[:,0]),
            __round(rgb[:,1]),
            __round(rgb[:,2])]

def color(array):
    col = {}
    col["red"] = [100, 20, 20] #suitable
    #During this time, embed the standard value of color
    col["heart"] = [100, 70, 70] #suitable
    
    max = 0
    result = ""
    for k, c in col.items:
       tmp = numpy.corrcoef(numpy.array(array), numpy.array(c))[0][1]
       if max < tmp:
           result = k
           max = tmp
    return result

def __round(array):
    return int(round(np.average(array)))

if __name__ == "__main__":
    '''
        xa :start point, xs :Block width, xb :end point
　　　　 ya :start point, ys :Block height, yb :end point
    '''
    pic = Image.open("/path/to/Image.png ", 'r')
    for i in xrange(6):
        for j in xrange(5):
           box = (xa + xs*i,
                  ya + ys*j,
                  xb + xs*i,
                  yb + ys*j)
           rgb = get_rgb(pic, box)
           print color(rgb)

In this case, there is no particular problem in terms of speed, and the detection rate is also reasonable. Occasionally, it drops like Zero division.

Conclusion

Isn't Puzzle & Dragons interesting? I've never done it.