calculate the transform required to transform a plate to match its edit ref.

calculate_inset_transform.py

"""
This is a cobbled together "copy paste" abomination which finds the bounding box of one image inside another.

Useful for matching compositions.
"""

import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt

def euclidean_distance(x1, x2):
    return np.sqrt(np.sum((x1 - x2)**2))


inset_img = cv.imread('/home/work/Downloads/image.png',cv.IMREAD_GRAYSCALE) # queryImage
orig_rgb = cv.imread('/home/work/Downloads/image(1).png')
orig_gray = cv.cvtColor(orig_rgb, cv.COLOR_BGR2GRAY)
# find the keypoints and descriptors with ORB
orb = cv.ORB_create()
kp1, des1 = orb.detectAndCompute(inset_img,None)
kp2, des2 = orb.detectAndCompute(orig_gray,None)

# find matching key points using brute force matching
bf = cv.BFMatcher(cv.NORM_HAMMING, crossCheck=True)
matches = bf.match(des1,des2)
matches = list(sorted(matches, key = lambda x:x.distance))


most_confident_matches = matches[:5]


scale_samples = []

origin_samples = []

for i in range(len(most_confident_matches)):
    for j in range(len(most_confident_matches)):
        if i == j:
            continue
        m1 = most_confident_matches[i]
        m2 = most_confident_matches[j]

        op1 = np.array(kp2[m1.trainIdx].pt)
        op2 = np.array(kp2[m2.trainIdx].pt)

        orig_distance = euclidean_distance(op1, op2)

        ip1 = np.array(kp1[m1.queryIdx].pt)
        ip2 = np.array(kp1[m2.queryIdx].pt)

        inset_distance = euclidean_distance(ip1, ip2)

        scale = orig_distance/inset_distance

        scale_samples.append(scale)

        origin = op1 - (ip1*scale)
        origin_samples.append(origin)

# resample the inset image to match the original image
scale = np.mean(scale_samples)
scaled_inset_img = cv.resize(inset_img, (0,0), fx=scale, fy=scale)

# find the rect of the scaled inset image in the original image.
# this could probably be done above but brain no math do.
res = cv.matchTemplate(orig_gray,scaled_inset_img, cv.TM_CCOEFF_NORMED)
min_val, max_val, min_loc, top_left = cv.minMaxLoc(res)

rect = (top_left[0], top_left[1], scaled_inset_img.shape[1], scaled_inset_img.shape[0])
cv.rectangle(orig_rgb, rect, (0,0,255), 2)




cv.imwrite("orig.png",orig_rgb)
cv.imwrite("inset.png",inset_img)