Color_Recognition_Kmeans.py

# -*- coding: utf-8 -*-
"""
Created on Tue Nov 14 15:51:57 2017

@author: twang

Using python and k-means to find the dominant colors in images

"""

import numpy as np
import cv2
import webcolors

def Color_Recognition_Kmeans(image_cropped):
        '''
        recognize the main color of the given box, input is box coordinate(x,y,w,h),return is color
        using kemans to cluster domianant color of a given box
        
        '''
        print('Color Recognition.....')
        
       # tic = time.time()
       
        #reshape image from [X][Y][3] to [X*Y,3]
        img_reshape = [img.reshape((-1,3)) for img in image_cropped]
        
        #reshape numpy 2D array to 1D array 
        pixels_flat = np.concatenate(img_reshape)
        
        #change to float 
        pixels_flat = np.float32(pixels_flat)
        
        # define criteria, number of clusters(K) and apply kmeans()
        # Define criteria = ( type, max_iter = 10 , epsilon = 1.0 )
        criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 10, 1.0)
        K = 3
        ret,label,center=cv2.kmeans(pixels_flat,K,None,criteria,10,cv2.KMEANS_RANDOM_CENTERS)
        
        #change to unit center
        center = np.uint8(center)
        
        #a list 
        center = [tuple([int(c) for c in color]) for color in center]
        
        #return unique class label of an array and its coming counts           
        l,c = np.unique(label,return_counts=True)
        
        #sorting list (l, c),reverse it from high to low 
        order = sorted(zip(c,l),reverse=True)
        
        #just show class label in order
        order = [l for c,l in order]

        #convert list to array in order, then to list
        center = np.asarray(center)[order].tolist()
        
        
        for index, RGB_value in enumerate(center):
            
            print (index, RGB_value)
            
            requested_colour =RGB_value
            
            try:
                
                closest_name = webcolors.rgb_to_name(requested_colour)
                
            except ValueError:

                min_colours = {}

                for key, name in webcolors.css3_hex_to_names.items():
                    r_c, g_c, b_c = webcolors.hex_to_rgb(key)
                    rd = (r_c - requested_colour[0]) ** 2
                    gd = (g_c - requested_colour[1]) ** 2
                    bd = (b_c - requested_colour[2]) ** 2
                    min_colours[(rd + gd + bd)] = name
                
                #actual_name = None
                closest_name = min_colours[min(min_colours.keys())]
                #closest_name = None
            
            print ( 'closest colour name:', closest_name)
            
            return closest_name  
            
# load the image
image = cv2.imread('test.jpg')

name = Color_Recognition_Kmeans(image)
cv2.putText(image,name,(5, 5), 2, 0.5, (0, 255, 0), 2, cv2.LINE_AA)
cv2.imshow("images", image)
cv2.waitKey(5000)

cv2.destroyAllWindows()