Algorithm 3.

RF-based Similarity Fusion Approach

1: Initially, consider the top $K$ images by applying similarity fusion (5) based on an equal feature weighting.

2: Obtain the user’s feedback about relevant images from the top $K$ images.

3: Calculate the new query vector ${\stackrel{\mathbf{ˆ}}{\mathbf{\text{f}}}}_q^F$ as the mean vector of the relevant images.

4: For each ranked list based on individual similarity matching, also consider top $K$ images and measure the effectiveness as $E\left({\mathbf{\text{f}}}^F\right)$ by utilizing equation (6).

5: Normalize the effectiveness or weight score to be in the range [0,1].

6: Utilize the normalized scores as updated weights in the similarity function of equation (7) for the final retrieval.

7: Continue, Steps 2 to 6 until no changes are noticed.