Script: homography estimation.py

#!/usr/bin/env python3
"""  Copyright (C) 2020 RidgeRun, LLC (http://www.ridgerun.com)
 All Rights Reserved.

 The contents of this software are proprietary and confidential to RidgeRun,
 LLC.  No part of this program may be photocopied, reproduced or translated
 into another programming language without prior written consent of
 RidgeRun, LLC.  The user is free to modify the source code after obtaining
 a software license from RidgeRun.  All source code changes must be provided
 back to RidgeRun without any encumbrance. """

""" Tool for estimate the homography matrix """

import argparse
import cv2
import json
import numpy as np
import sys

HOMOGRAPHY_DIMENSION = 3
MIN_MATCHES = 4

def drawMatches(imageA, imageB, kpsA, kpsB, matches, status):
    '''
   Returns an image with the matches found.

           Parameters:
                   imageA (np.array): Image A
                   imageB (np.arry): Image B
                   kpsA (np.arry): Keypoints for image A
                   kpsB (np.arry): Keypoints for image B
                   matches (np.array): List of found matches between image A and image B
                   status (int): status of homography estimation

           Returns:
                   Image
   '''

    # initialize the output visualization image
    (hA, wA) = imageA.shape[:2]
    (hB, wB) = imageB.shape[:2]
    vis = np.zeros((max(hA, hB), wA + wB, 3), dtype="uint8")
    vis[0:hA, 0:wA] = imageA
    vis[0:hB, wA:] = imageB

    for ((trainIdx, queryIdx), s) in zip(matches, status):
        # only process the match if the keypoint was successfully
        # matched
        if s == 1:
            # draw the match
            ptA = (int(kpsA[queryIdx][0]), int(kpsA[queryIdx][1]))
            ptB = (int(kpsB[trainIdx][0]) + wA, int(kpsB[trainIdx][1]))
            cv2.line(vis, ptA, ptB, (0, 255, 0), 1)

    return vis


def detectAndDescribe(gray, mask):
    '''
    Returns the keyponts and the correspoinding descriptors for the
    gray image.

            Parameters:
                    gray (np.array): Input image in grayscale.
                    mask (np.arry): Mask to apply before the feature extraction.

            Returns:
                    Tuple with the keypoins and their descriptors.
    '''

    # detect and extract features from the image
    detector = cv2.xfeatures2d.SIFT_create()
    (kps, features) = detector.detectAndCompute(gray, mask)
    kps = np.float32([kp.pt for kp in kps])

    return (kps, features)


def matchKeypoints(kpsA, kpsB, featuresA, featuresB,
                   ratio, reprojThresh):
    '''
    Function that search for the correspondiencies for the given descriptors (featuresA and featuresB) and
    with those correspondencies keypoints, the homography matrix is calculated.

            Parameters:
                    kpsA (np.array): Keypoints of image A.
                    kpsB (np.arry): Keypoints of image B.
                    featuresA (np.arry): Descriptors of image A
                    featuresB (np.arry): Descriptors of image B
                    ratio (float): Max distance between a possible correspondence.
                    reprojThresh (float): Reprojection error of the homography estimation

            Returns:
                    Tuple with the resulting matches, the homography H and the estimation status.
    '''

    # compute the raw matches and initialize the list of actual matches
    matcher = cv2.DescriptorMatcher_create("BruteForce")
    rawMatches = matcher.knnMatch(featuresA, featuresB, 2)
    matches = []

    # loop over the raw matches
    for m in rawMatches:
        # ensure the distance is within a certain ratio of each
        # other (i.e. Lowe's ratio test)
        if len(m) == 2 and m[0].distance < m[1].distance * ratio:
            matches.append((m[0].trainIdx, m[0].queryIdx))

    print("matches: " + str(len(matches)))
    if len(matches) >= MIN_MATCHES:
        # construct the two sets of points
        ptsA = np.float32([kpsA[i] for (_, i) in matches])
        ptsB = np.float32([kpsB[i] for (i, _) in matches])

        # compute the homography between the two sets of points
        (H, status) = cv2.findHomography(ptsA, ptsB, cv2.RANSAC,
                                         reprojThresh)

        # return the matches along with the homograpy matrix
        # and status of each matched point
        return (matches, H, status)

    # otherwise, no homograpy could be computed
    return None


def imageEnhancement(img, sigma):
    '''
    Rerturns an image with some enhancement applied. Remove the noise of the input image img
    using a Gaussian filter and then convert it to grayscale.

            Parameters:
                    img (np.array): Input image img
                    sigma (float): Sigma value of the Gaussian filter.

            Returns:
                    Image
    '''

    # Declare the variables we are going to use
    kernel_size = 5

    # Remove noise
    img = cv2.GaussianBlur(img, (kernel_size, kernel_size), sigma)

    # Convert to grayscale
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

    return gray


def removeDistortion(img, K, d):
    '''
    Returns an image with the distortion provoked by the camera lens.

            Parameters:
                    img (np.array): Image img.
                    K (np.arry): Camera matrix with a dimension of (3x3).
                    d (np.arry): Vector with the distortion coefficients.
            Returns:
                    Undistorted image.
    '''

    # Read an example image and acquire its size
    h, w = img.shape[:2]

    # Generate new camera matrix from parameters
    newcameramatrix, roi = cv2.getOptimalNewCameraMatrix(K, d, (w, h), 0)

    # Generate look-up tables for remapping the camera image
    mapx, mapy = cv2.initUndistortRectifyMap(
        K, d, None, newcameramatrix, (w, h), 5)

    # Remap the original image to a new image
    newimg = cv2.remap(img, mapx, mapy, cv2.INTER_LINEAR)
    return newimg


def parseJSON(filename):
    '''
    Returns the algorithm variables read from a JSON configuration file.

            Parameters:
                    filename (string): Path to the JSON configuration file.

            Returns:
                    Tuple the the read variables: (K, D, reprojError, matchRatio, sigma, overlap, crop, fov, undistort)
    '''

    with open(filename) as json_file:
        data = json.load(json_file)

        # load config variables
        reprojError = float(data['reprojError'])
        matchRatio = float(data['matchRatio'])
        sigma = float(data['sigma'])
        fov = float(data['fov'])
        overlap = float(data['overlap'])
        crop = float(data['crop'])
        undistort = bool(data['undistort'])

        # load camera matrix and distortion coeficients
        K = None
        D = None
        if undistort:
            K = np.zeros(HOMOGRAPHY_DIMENSION * HOMOGRAPHY_DIMENSION)
            for i, k in enumerate(data['K']):
                K[i] = k
            K = K.reshape(HOMOGRAPHY_DIMENSION, HOMOGRAPHY_DIMENSION)

            D = np.zeros([1, len(data['d'])])
            for i, d in enumerate(data['d']):
                D[0, i] = d

        return (K, D, reprojError, matchRatio,
                sigma, overlap, crop, fov, undistort)

def left_fixed(config, targetImage, originalImage, homographyScale):
    '''
    Performs the homography estimation betwen two images, leaving the left one fixed and
    transforming the right one to align them.

            Parameters:
                    config (string): Path to the JSON configuration file.
                    targetImage (string): Path to the traget iamge.
                    originalImage (string): Path to the original image.
                    homographyScale (float): Scale factor for the generated homography.

            Returns:
                    No return value
    '''

    (K, d, reprojThresh, ratio, sigma, overlap,
     crop, fov, undistort) = parseJSON(config)

    # Load images
    target = cv2.imread(targetImage)
    original = cv2.imread(originalImage)
    height, width = target.shape[:2]

    # crop
    cropImg = int((crop * target.shape[1]) / fov)
    if cropImg > 0:
        target = target[:, :-1 * cropImg]
        target = cv2.resize(target,(width, height), interpolation = cv2.INTER_CUBIC)
        original = original[:, cropImg:]
        original = cv2.resize(original,(width, height), interpolation = cv2.INTER_CUBIC)

    # remove undistort
    if undistort:
        target = removeDistortion(target, K, d)
        original = removeDistortion(original, K, d)

    # Enhance images
    enhancedTarget = imageEnhancement(target, sigma)
    enhancedOriginal = imageEnhancement(original, sigma)

    # Mask
    overlapImg = int((overlap * target.shape[1]) / fov)
    targetMask = np.zeros(enhancedTarget.shape, dtype=np.uint8)
    targetMask[0:, (-1 * overlapImg):] = 1
    originalMask = np.zeros(enhancedOriginal.shape, dtype=np.uint8)
    originalMask[0:, 0:overlapImg] = 1

    # Extract keypoints
    (kpsA, featuresA) = detectAndDescribe(enhancedOriginal, originalMask)
    (kpsB, featuresB) = detectAndDescribe(enhancedTarget, targetMask)

    # Match features
    (matches, H, status) = matchKeypoints(
        kpsA, kpsB, featuresA, featuresB, ratio, reprojThresh)

    vis = drawMatches(original, target, kpsA, kpsB, matches, status)
    cv2.imwrite('vis.jpg', vis)

    # Apply homography
    result = cv2.warpPerspective(
        original, H, (target.shape[1] + original.shape[1], original.shape[0]))
    result[0:target.shape[0], 0:target.shape[1]] = target
    cv2.imwrite('result.jpg', result)

    # Scale homography
    if homographyScale > 0:
        Hmap = cv2.reg_MapProjec(H)
        Hmap.scale(homographyScale)
        H = Hmap.getProjTr()

    print("""RC_HOMOGRAPHY=\"{{\\"h00\\":{},\\"h01\\":{}, \\"h02\\":{}, \\"h10\\":{}, \\"h11\\":{}, \\"h12\\":{}, \\"h20\\":{}, \\"h21\\":{}, \\"h22\\":{}}}\"""".format(
          H[0][0],
          H[0][1],
          H[0][2],
          H[1][0],
          H[1][1],
          H[1][2],
          H[2][0],
          H[2][1],
          H[2][2])
          )


def right_fixed(config, targetImage, originalImage, homographyScale):
    '''
    Performs the homography estimation betwen two images, leaving the right one fixed and
    transforming the left one to align them.

            Parameters:
                    config (string): Path to the JSON configuration file.
                    targetImage (string): Path to the traget iamge.
                    originalImage (string): Path to the original image.
                    homographyScale (float): Scale factor for the generated homography.

            Returns:
                    No return value
    '''

    (K, d, reprojThresh, ratio, sigma, overlap,
     crop, fov, undistort) = parseJSON(config)

    # Load images
    target = cv2.imread(targetImage)
    original = cv2.imread(originalImage)
    height, width = target.shape[:2]

    # crop
    cropImg = int((crop * target.shape[1]) / fov)
    if cropImg > 0:
        target = target[:, cropImg:]
        target = cv2.resize(target,(width, height), interpolation = cv2.INTER_CUBIC)
        original = original[:, :-1 * cropImg]
        original = cv2.resize(original,(width, height), interpolation = cv2.INTER_CUBIC)

    # remove undistort
    if undistort:
        target = removeDistortion(target, K, d)
        original = removeDistortion(original, K, d)

    # Enhance images
    enhancedTarget = imageEnhancement(target, sigma)
    enhancedOriginal = imageEnhancement(original, sigma)

    # Mask
    overlapImg = int((overlap * target.shape[1]) / fov)
    targetMask = np.zeros(enhancedTarget.shape, dtype=np.uint8)
    targetMask[0:, 0:overlapImg] = 1
    originalMask = np.zeros(enhancedOriginal.shape, dtype=np.uint8)
    originalMask[0:, (-1 * overlapImg):] = 1

    # Extract keypoints
    (kpsA, featuresA) = detectAndDescribe(enhancedOriginal, originalMask)
    (kpsB, featuresB) = detectAndDescribe(enhancedTarget, targetMask)

    # Match features
    (matches, H, status) = matchKeypoints(
        kpsB, kpsA, featuresB, featuresA, ratio, reprojThresh)
    H = np.linalg.inv(H)

    # X translation shift
    H[0][-1] += original.shape[1]

    vis = drawMatches(target, original, kpsB, kpsA, matches, status)
    cv2.imwrite('vis.jpg', vis)

    # Apply homography
    result = cv2.warpPerspective(
        original, H, (target.shape[1] + original.shape[1], original.shape[0]))
    result[0:target.shape[0], target.shape[1]:] = target
    cv2.imwrite('result.jpg', result)

    # Scale homography
    if homographyScale > 0:
        Hmap = cv2.reg_MapProjec(H)
        H[0][-1] -= original.shape[1]
        Hmap.scale(homographyScale)
        H = Hmap.getProjTr()

    print("""LC_HOMOGRAPHY=\"{{\\"h00\\":{},\\"h01\\":{}, \\"h02\\":{}, \\"h10\\":{}, \\"h11\\":{}, \\"h12\\":{}, \\"h20\\":{}, \\"h21\\":{}, \\"h22\\":{}}}\"""".format(
          H[0][0],
          H[0][1],
          H[0][2],
          H[1][0],
          H[1][1],
          H[1][2],
          H[2][0],
          H[2][1],
          H[2][2])
          )

def cmdline(argv):
    '''
    Function taht parse the command line optins before execute the algorithm

            Parameters:
                    config (list): List of command line arguments.

            Returns:
                    No return value
    '''

    prog = argv[0]
    parser = argparse.ArgumentParser(
        prog=prog,
        description='Tool for use the prediction capabilities of the models in the Adversarial Anomaly Detector.',
        epilog='Type "%s <command> -h" for more information.' % prog)

    subparsers = parser.add_subparsers(dest='command')
    subparsers.required = True

    def add_command(cmd, desc, example=None):
        epilog = 'Example: %s %s' % (
            prog, example) if example is not None else None
        return subparsers.add_parser(
            cmd, description=desc, help=desc, epilog=epilog)

    p = add_command('left_fixed', 'Estimation of homography between two images')

    p.add_argument('--config',          help='Path of configure file', default='')
    p.add_argument('--targetImage',     help='Path of the target image', default='')
    p.add_argument('--originalImage',   help='Path of the original image', default='')
    p.add_argument('--homographyScale', help='Scale factor of the homography. For example if you go from 1920x1080 in the estimation to 640x360 in the processing the scale factor should be 1/3', type=float, default=0)

    p = add_command('right_fixed', 'Estimation of homographies two images')

    p.add_argument('--config',          help='Path of configure file', default='')
    p.add_argument('--targetImage',     help='Path of the target image', default='')
    p.add_argument('--originalImage',   help='Path of the original image', default='')
    p.add_argument('--homographyScale', help='Scale factor of the homography. For example if you go from 1920x1080 in the estimation to 640x360 in the processing the scale factor should be 1/3', type=float, default=0)

    args = parser.parse_args(argv[1:] if len(argv) > 1 else ['-h'])
    func = globals()[args.command]
    del args.command
    func(**vars(args))

# ----------------------------------------------------------------------------


if __name__ == "__main__":
    cmdline(sys.argv)

# ----------------------------------------------------------------------------