Add homography

OCR/image-alignment/run-homography.py

@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+
+"""
+Uses quadrilaterial edge detection and executes a four-point perspective transform on a source image.
+"""
+
+import numpy as np
+import cv2 as cv
+
+img1 = cv.imread('pertussis-template.png',cv.IMREAD_GRAYSCALE) # queryImage
+img2 = cv.imread('output/pertussis-filled-2.png',cv.IMREAD_GRAYSCALE) # trainImage
+
+ratio = 0.5
+
+img1 = cv.resize(img1, None, fx=ratio, fy=ratio, interpolation=cv.INTER_LANCZOS4)
+img2 = cv.resize(img2, None, fx=ratio, fy=ratio, interpolation=cv.INTER_LANCZOS4)
+cv.imwrite('s1.png', img1)
+cv.imwrite('s2.png', img2)
+
+sift = cv.SIFT_create()
+
+# find the keypoints and descriptors with SIFT
+kp1, descriptors1 = sift.detectAndCompute(img1,None)
+kp2, descriptors2 = sift.detectAndCompute(img2,None)
+
+matcher = cv.DescriptorMatcher_create(cv.DescriptorMatcher_FLANNBASED)
+knn_matches = matcher.knnMatch(descriptors1, descriptors2, 2)
+
+# import pdb; pdb.set_trace()
+
+# Filter matches using the Lowe's ratio test
+# use an aggressive threshold here- the larger the image the more aggresively this should be filtered
+ratio_thresh = 0.3
+good_matches = []
+for m,n in knn_matches:
+    if m.distance < ratio_thresh * n.distance:
+        good_matches.append(m)
+
+img3 = cv.drawMatches(img1, kp1, img2, kp2, good_matches, None, flags=cv.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)
+cv.imwrite('test-match.png', img3)
+
+src_pts = np.float32([ kp1[m.queryIdx].pt for m in good_matches ]).reshape(-1,1,2)
+dst_pts = np.float32([ kp2[m.trainIdx].pt for m in good_matches ]).reshape(-1,1,2)
+
+M, mask = cv.findHomography(dst_pts, src_pts, cv.RANSAC,5.0)
+matchesMask = mask.ravel().tolist()
+
+h,w = img1.shape
+pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
+dst = cv.perspectiveTransform(pts,M)
+out = cv.warpPerspective(img2, M, (w, h))
+
+cv.imwrite('test-transform.png', out)