Enhancing Sketch-Based Image Retrieval by Re-Ranking and Relevance - - PowerPoint PPT Presentation
Enhancing Sketch-Based Image Retrieval by Re-Ranking and Relevance Feedback Heechan Shin CS688 Student paper presentation Enhancing Sketch-Based Image Retrieval by Re-Ranking and Relevance Feedback ( IEEE TIP 16 ) 2018-12-03 2 Contents
2018-12-03
“Enhancing Sketch-Based Image Retrieval by Re-Ranking and Relevance Feedback” ( IEEE TIP 16 )
2
3
What a user want to find What a user queries
How to measure the relevance of an image and a query sketch?
4
Angular Radial Partitioning(ARP) Edgel index
Edgel index : Cao, Yang, et al. "Edgel index for large-scale sketch-based image search." (2011): 761-768. ARP : Chalechale, Abdolah, Alfred Mertins, and G. Naghdy. "Edge image description using angular radial partitioning." IEE Proceedings-Vision, Image and Signal Processing 151.2 (2004): 93-101.
5
Image Partitioning Pixels in each partition Fourier transformed
6
7
8
Any SBIR Optimizing module
Initial result Final result
9
relevant images using contours
10
relevant images using contours
11
Initial result (size N) Select images (size R) (R < N) Find near-duplicated images using existing image matching approach (ex, binary edge-SIFT) Cluster near-duplicated images (size of cluster K) (K <= R) Relevant group
Rank high
12
Ranked images (size N)
Top ranked image (Standard image, )
Calculate similarity score to Similarity score = , , = 1~ (ex. = , = 1.0) ( , ) ( , ) ( , ) ( , ) 1 2 3 N … Re-ranked images according to (size N) Select top M images
13
= 2 − Σ
+ = 2,
, = Σ
14
Re-ranked images (size M) … Create contour from image (size M)
Relevant Feedback Score ∶ Σ
, × ; = 1, … ,
ℎ
New query New query New query
15
, × ; = 1, … ,
: Initial score of image
,
+ × ; = 1, … ,
16
relevant images using contours
17
( , ) ( , ) ( , ) ( , )
1 2 3 N
… …
New query New query New query
( , ) ( , ) ( , ) ( , )
1 2 3 N
…
18
sketch-describable images + 523 sketches
19
Result of authors’ dataset Result of SBIR_100K dataset
20
+1.28s +0.91s
21