1.背景

项目需要，透过可视化的图片连接两张图的相似点检查SIFT之正确率

2.效果

3.代码

def drawMatchLine(Img1Path , Img2Path ):
    # 读入图片    
    Img1 = cv2.imread(Img1Path)
    Img2 = cv2.imread(Img2Path)
    
    # 创建识别器
    detector = cv2.xfeatures2d.SIFT_create(200)
    
    # 计算KeyPoints
    kp1 , des1 = detector.detectAndCompute(Img1,None)
    kp2 , des2 = detector.detectAndCompute(Img2,None)
    
    # BFMatcher
    bf = cv2.BFMatcher()
    matchers = bf.knnMatch(des1 , des2 , k=2)
    
    # 相似列表
    Match = []
    for m,n in matchers:
        if m.distance <  0.50*n.distance:
            Match.append(m)
            
    # 查看两张图片的宽及高
    height1 , width1 = Img1.shape[:2]
    height2 , width2 = Img2.shape[:2]
    
    # 像素调整
    vis = np.zeros((max(height1, height2), width1 + width2, 3), np.uint8)
    vis[:height1, :width1] = Img1
    vis[:height2, width1:width1 + width2] = Img2
    
    p1 = [kpp.queryIdx for kpp in Match[:20]]
    p2 = [kpp.trainIdx for kpp in Match[:20]]
    
    
    
    post1 = np.int32([kp1[pp].pt for pp in p1])
    post2 = np.int32([kp2[pp].pt for pp in p2]) + (width1, 0)
    
    
    for (x1, y1), (x2, y2) in zip(post1, post2):
        cv2.line(vis, (x1, y1), (x2, y2), (0,0,255))
        
    
    cv2.namedWindow("match",cv2.WINDOW_NORMAL)
    cv2.imshow("match", vis)    
    cv2.waitKey(0)
    cv2.destroyAllWindows()

4.后续

SIFT算法列出的特征点要再对其阀域做探讨，虽然保存特征点的文件大小不大，但若需要遍历大量资料库时仍需考虑特征点之准确及数量。

有问题欢迎留言讨论指教，谢谢！