gradientDescents2/Panorama

1import cv22import numpy as np3 4def computeDes(img1, img2):5 6    # Convert images to grayscale7    gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)8    gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)9 10    # Initialize the feature detector and extractor (e.g., SIFT)11    sift = cv2.SIFT_create()12 13    # Detect keypoints and compute descriptors for both images14    keypoints1, descriptors1 = sift.detectAndCompute(gray1, None)15    keypoints2, descriptors2 = sift.detectAndCompute(gray2, None)16 17    return (keypoints1, descriptors1), (keypoints2, descriptors2)18 19def feature_matching(keypoints1, descriptors1, keypoints2, descriptors2):20 21    # Initialize the feature matcher using FLANN matching22    index_params = dict(algorithm=0, trees=5)23    search_params = dict(checks=50)24    flann = cv2.FlannBasedMatcher(index_params, search_params)25 26    # Select the top N matches27    num_matches = 5028    # Match descriptors using FLANN29    matches_flann = flann.match(descriptors1, descriptors2)30 31    # Sort the matches by distance (lower is better)32    matches= sorted(matches_flann, key=lambda x: x.distance)[:num_matches]33 34    # Draw the top N matches35    # image_matches_flann = cv2.drawMatches(image1, keypoints1, image2, keypoints2, matches_flann[:num_matches], None)36 37    # Extract matching keypoints38    src_points = np.float32([keypoints1[match.queryIdx].pt for match in matches]).reshape(-1, 1, 2)39    dst_points = np.float32([keypoints2[match.trainIdx].pt for match in matches]).reshape(-1, 1, 2)40 41    return src_points, dst_points42 43def findHomography(src_points, dst_points):44    homography, _ = cv2.findHomography(src_points, dst_points, cv2.RANSAC, 5.0)45    return homography46 47def stitImage(right_img, left_img):48 49    (keypoints1, descriptors1), (keypoints2, descriptors2) = computeDes(right_img, left_img)50 51    src_points, dst_points = feature_matching(keypoints1, descriptors1, keypoints2, descriptors2)52 53    homography = findHomography(src_points, dst_points)54 55    h1, w1 = right_img.shape[:2]56    h2, w2 = left_img.shape[:2]57    result = cv2.warpPerspective(right_img, homography, (w1+w2, max(h1,h2)))58    # result[0:h2, 0:w2] = left_img59    result = blend_imgs(left_img, result, 50)60    # cv2.imwrite("r2.jpg", result)61 62    result = crop(result)63 64    return result65 66def blend_imgs(img1, img2, width_overlap):67    h1, w1 = img1.shape[:2]68    h2, w2 = img2.shape[:2]69    for y in range(h1):70        for x in range(w1):71            if x >= (w1 - width_overlap):72                img1[y,x] = img1[y,x]*((w1-x)/width_overlap)73    74    for y in range(h2):75        for x in range(w1):76            if x >= (w1 - width_overlap):77                img2[y,x] = img2[y,x]*((x+width_overlap - w1)/width_overlap)78    79    img2[:,w1 - width_overlap: w1] += img1[:,-width_overlap:]80 81    img2[0:h1, 0:w1 - width_overlap] = img1[:,0:w1 - width_overlap]82 83    return img284 85def crop(img):86 87    # if img.dtype == np.float64:88    #     img = (img * 255).astype(np.uint8)89 90    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)91 92    # Create binary mask to identify dest image93    _, binary_mask = cv2.threshold(gray, 1, 255, cv2.THRESH_BINARY)94 95    # Find contours around dest image96    contours, _ = cv2.findContours(binary_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)97 98    # Find the min boundary 99    x, y, w, h = cv2.boundingRect(contours[0])100 101    # crop image102    cropped_image = img[y:y+h, x:x+w]103 104    return cropped_image105 106l_img = cv2.imread("images/left.jpg")107r_img = cv2.imread("images/right.jpg")108 109if __name__ == "__main__":110 111    result = stitImage(r_img, l_img)112    cv2.imshow("Stit Image", result)113    cv2.waitKey(0)114    cv2.destroyAllWindows()