09-colorize.py

import numpy 
import cv2 
numpyInput = cv2.imread(filename='./samples/prokudin.png', flags=cv2.IMREAD_GRAYSCALE).astype(numpy.float32)/255.0 
# align the individual images such that their combination results in a proper color image 
# splitting the input into the three individual channels while slightly corrping the boundary 
intFirst, intSecond = int(1.0 * numpyInput.shape[0] / 3.0), int(2.0 * numpyInput.shape[0] / 3.0) 
numpyB = numpyInput[:intFirst, :][50:-50, 50:-50] 
numpyG = numpyInput[intFirst:intSecond, :][50:-50, 50:-50] 
numpyR = numpyInput[intSecond:, :][50:-50, 50:-50] 
# find the homography matrices between the images using cv2.findTransformECC 
# based on these matrices, warp numpyG and numpyR towards numpyB using cv2.warpPerspective 
# make sure to use inverse warping as stated in the docuḿentation of cv2.findTransformECC 
# once the channels are aligned, they can simply be stacked to create the color image 
numpyHomographyG = cv2.findTransformECC(templateImage=numpyB, inputImage=numpyG, warpMatrix=numpy.eye(3, 3, 0, numpy.float32), motionType=cv2.MOTION_HOMOGRAPHY)[1]
numpyHomographyR = cv2.findTransformECC(templateImage=numpyB, inputImage=numpyR, warpMatrix=numpy.eye(3, 3, 0, numpy.float32), motionType=cv2.MOTION_HOMOGRAPHY)[1] 
numpyG = cv2.warpPerspective(src=numpyG, M=numpyHomographyG, dsize=(numpyG.shape[1], numpyG.shape[0]), flags=cv2.INTER_LINEAR + cv2.WARP_INVERSE_MAP) 
numpyR = cv2.warpPerspective(src=numpyR, M=numpyHomographyR, dsize=(numpyR.shape[1], numpyR.shape[0]), flags=cv2.INTER_LINEAR + cv2.WARP_INVERSE_MAP) 
numpyOutput = numpy.stack([numpyB, numpyG, numpyR], 2)
cv2.imwrite(filename='./09-colorize.png', img=(numpyOutput * 255.0).clip(0.0, 255.0).astype(numpy.uint8))