使用opencv python检测SURF徽标

Question

I am trying to detect some logos of known software related things in pictures. I use Opencv 2.4.5 with python 2.7. I would like to use SURF detector implemented in opencv but the problem is that I don't obtain good results. There are lots of false negative and false positive. My code is :

import cv2
import numpy as np

def detectLogo(template, img):

    templateg = cv2.cvtColor(template, cv2.COLOR_BGR2GRAY)
    imgg = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

    # SURF extraction
    hessian_threshold = 30
    surf = cv2.SURF(hessian_threshold)
    kp, desc = surf.detect(imgg, None, useProvidedKeypoints = False)

    # KNN
    samples = np.array(desc)
    responses = np.arange(len(kp), dtype = np.float32)
    knn = cv2.KNearest()
    knn.train(samples, responses)

    # Loading template and searching for similar kp
    kp2, desc2 = surf.detect(templateg, None, useProvidedKeypoints = False)

    matched = 0
    total = 0
    for h,des in enumerate(desc2):
        des = np.array(des,np.float32).reshape((1,128))
        retval, results, neigh_resp, dists = knn.find_nearest(des,1)
        res,dist =  int(results[0][0]),dists[0][0]
        total += 1
        if dist<0.1: # draw matched keypoints in red color
            color = (0,0,255)
            matched += 1
        else:
            color = (255,0,0)
        #Draw matched key points on original image
        x,y = kp[res].pt
        center = (int(x),int(y))
        cv2.circle(img,center,2,color,-1)

        #Draw matched key points on template image
        x,y = kp2[h].pt
        center = (int(x),int(y))
        cv2.circle(template,center,2,color,-1)

    cv2.imwrite("../resources/template.jpg", template)
    cv2.imwrite("../resources/image.jpg", img)
    return matched / float(total)

template = cv2.imread("../resources/pictures/appleLogo.jpg")
img = cv2.imread("../resources/pictures/pic2.jpg")
print detectLogo(template, img)

Below are results.

_{(source: hostingpics.net )}

Template url .

_{(source: hostingpics.net )}

Image url .

Matched points don't correspond at all to the loho and I got same results for two completely different images.

I think it's the only solution to perform this task but where is the problem with this dectection ? Thank you in advance.

Alexandre

Answer 1

Interest points along edges get suppressed because they are not invariant in scale space. Since your logo is completely defined by edges, an intertest point detector that works like SURF (or SIFT) is not going to work very well. Probably training up a Haar classifier, like the Viola-Jones one in OpenCV, would work better.