Hi,

I would like to understand well the "Manifold-Ranking Based Image Retrieval" principale espcially using positive and negative samples. 

Knowing that the images of the database are classified in different classes.

According to me, Manifold-Ranking Based Image Retrieval algorithm is as follows: 

1.assign postive ranking score to the query (1) and zero to the remainng points of the Image database.

2.generate a weighted graph:

            -Compute K nearest neighbors for each image

            -Connect 2 images with edge if they are neighbors

            -Form the  affinity matrix to define the edge weights

            -Normalise the  affinity matrix

3. spread the score of all images to the neighbors via the weighted graph until the images (which have ranking score Zero  ) get ranking score  different from Zero.

4. rank the images according to their ranking scores (largest ones in the top)

Now, I present Manifold-Ranking Based Image Retrieval using relevance feedback (positive and negative images) algorithm

5. Active leanrning selects the positive and negative images.

6. Rerun Manifold-Ranking Based Image Retrieval algorithm

Q.1 The stated algorithms are correct?

Q.2 When I execute different queries for the same class, I get the same retrieved images but with different ranking score for each query . Could you explain why?

Q.3 How to compute the effectiveness of the image retrieval?

Attached, the concerned paper.

Thank you a lot.

Best Regards.

Khadidja Belattar.

More Khadidja Belattar's questions See All
Similar questions and discussions