Spaces:
Sleeping
Sleeping
File size: 4,340 Bytes
cd35f72 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 |
import cv2
import numpy as np
import gradio as gr
def sift_ransac_matching(image, template, ratio_thresh=0.75, ransac_reproj_thresh=5.0, score_thresh=0.8, min_inliers=10):
"""
Returns:
- result_text (str)
- match_score (float, inlier_ratio = inliers / good_matches)
- detected (bool)
- annotated_image (numpy array, BGR)
"""
if image is None or template is None:
return "Invalid input images.", 0.0, False, image
# Convert to grayscale
gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
gray_template = cv2.cvtColor(template, cv2.COLOR_BGR2GRAY)
# Initialize SIFT detector
sift = cv2.SIFT_create()
# Find keypoints and descriptors
kp_img, des_img = sift.detectAndCompute(gray_image, None)
kp_tmpl, des_tmpl = sift.detectAndCompute(gray_template, None)
if des_img is None or des_tmpl is None or len(kp_img) == 0 or len(kp_tmpl) == 0:
return "No features detected. Template not found.", 0.0, False, image
# KNN matches with Lowe's ratio test
bf = cv2.BFMatcher()
knn = bf.knnMatch(des_img, des_tmpl, k=2)
good_matches = []
for pair in knn:
if len(pair) < 2:
continue
m, n = pair
if m.distance < ratio_thresh * n.distance:
good_matches.append(m)
if len(good_matches) < 4:
return f"Not enough good matches ({len(good_matches)}). Template not found.", 0.0, False, image
# Build point arrays
src_pts = np.float32([kp_img[m.queryIdx].pt for m in good_matches]).reshape(-1, 1, 2) # image -> src
dst_pts = np.float32([kp_tmpl[m.trainIdx].pt for m in good_matches]).reshape(-1, 1, 2) # template -> dst
# Homography with RANSAC
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, ransac_reproj_thresh)
if M is None or mask is None:
return "Homography failed. Template not found.", 0.0, False, image
inliers = int(mask.ravel().sum())
match_score = float(inliers) / float(len(good_matches)) # inlier ratio
# Detection decision (both quality and absolute inlier count can help stability)
detected = (match_score >= score_thresh) and (inliers >= min_inliers)
# Annotate image by projecting template corners back to image space
annotated = image.copy()
if detected:
try:
# M maps image -> template; invert to map template -> image
Minv = np.linalg.inv(M)
h, w = gray_template.shape[:2]
tmpl_corners = np.float32([[0, 0], [w, 0], [w, h], [0, h]]).reshape(-1, 1, 2)
proj = cv2.perspectiveTransform(tmpl_corners, Minv)
proj = proj.astype(int)
# Draw polygon
cv2.polylines(annotated, [proj.reshape(-1, 2)], True, (0, 255, 0), 2)
# Put label
txt = f"Detected | score={match_score:.3f} | inliers={inliers}/{len(good_matches)}"
x, y = proj.reshape(-1, 2)[0]
cv2.putText(annotated, txt, (max(0, x), max(20, y)),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2)
except np.linalg.LinAlgError:
# If inversion fails, still return score/decision
pass
result_text = (
f"Template {'found' if detected else 'not found'} | "
f"score={match_score:.3f} | inliers={inliers}/{len(good_matches)} good matches"
)
return result_text, match_score, detected, annotated
# Gradio UI
iface = gr.Interface(
fn=sift_ransac_matching,
inputs=[
gr.Image(type="numpy", label="Target Image"),
gr.Image(type="numpy", label="Template Image"),
gr.Slider(0.5, 0.95, value=0.75, step=0.01, label="Lowe ratio threshold"),
gr.Slider(1.0, 10.0, value=5.0, step=0.5, label="RANSAC reprojection threshold"),
gr.Slider(0.1, 1.0, value=0.8, step=0.01, label="Match score threshold"),
gr.Slider(0, 100, value=10, step=1, label="Minimum inliers"),
],
outputs=[
gr.Text(label="Result"),
gr.Number(label="Match score (inlier ratio)"),
gr.Checkbox(label="Detected?"),
gr.Image(label="Annotated result"),
],
title="SIFT + RANSAC Template Search",
description="Uploads a target and a template. Shows detection decision, match score, and an annotated result."
)
if __name__ == "__main__":
iface.launch()
|