4. Introduction
Target search in content-based image retrieval (CBIR) systems refers to finding a
specific (target) image such as a particular registered logo or a specific historical
photograph.
Existing techniques, designed around query refinement based on relevance feedback
(RF), suffer from slow convergence, and do not guarantee to find intended targets.
To address these limitations, we propose several efficient query point movement
methods.
We prove that our approach is able to reach any given target image with fewer
iterations in the worst and average cases.
We propose a new index structure and query processing technique to improve
retrieval effectiveness and efficiency.
We also consider strategies to minimize the effects of users’ inaccurate RF.
5. Purpose
The main purpose of this document is to meet the requirements as
mentioned in the SRS.
Develop a CBIR system that focuses on target search techniques, and
faster than the existing CBIR system and which is not a victim to slow
convergence, local maximum traps, minimizing the resource requirements.
CBIR system that can handle inefficient relevance feedback (RF).
The user is provided with a flexible user interface in which he/she has to
login to the system to use the software.
After login process, the user presents the image of similarity to search, by
browsing the local computer.
The users’ query is processed and a list of relevant images are produced.
6. Purpose
The user picks the images as positive and negative and the positive images
are considered for next round of retreival.
7. Scope
filtering and law enforcement markets .
Crime detection
Cencoring
Some benefits
1. User Feedback is included.
2. Reduces the unrelated searches.
3. The software is sensitive to inaccurate feedback.
4. Future retrievals of images can be processed faster.
8. Scope
5. Guarantees that the image is found.
6. Can reach target image with fewer iterations.
7. The scenario of local maximum traps and slow convergence is totally
eradicated.
8. The images are searched using image properties.
9. The system is not sensitive to users’ inaccurate relevance feedback.
12. Algorithm
1) Naïve random scan (NRS) method
I
The NRS method randomly retrieves k different images at a time until the user
finds the target image or the remaining set is exhausted.
At each iteration, a set of k random images are retrieved from the candidate
(i.e.unchecked) set S’ for relevance feedback , and S’ is then reduced by k .
In the best case, NRS takes one iteration Ω (1).
while the worst case requires S/K iterations.
At each iteration, a set of k random images are retrieved from the candidate
(i.e.unchecked) set S’ for relevance feedback , and S’ is then reduced by k .
In the best case, NRS takes one iteration Ω (1).
while the worst case requires S/K iterations.
13. Algorithm
2)Local neighboring movement (LNM) method
LNM is similar to NRS except for steps 5 step 6 which is explained as follows:
step5: Qr ←<nQ,PQ,WQ,DQ, S’,k> based on the user’srelevance feedback.
step6: Sk ← EVALUATEQUERY(Qr) /* perform a constrained k-NN query */
Qr is constructed such that it moves towards neighboring relevant points and away
from irrelevant ones, and a query is now evaluated against S’ instead of S.
One iteration is required in the best case Ω(1).
The worst case O(1) is given by
the average case o(1) is given by .
14. Algorithm
3) Neighboring divide and conquer (NDS) method
Voronoi diagrams in NDC to reduce search space.
The Voronoi diagram approach finds the nearest neighbors of a given query point
by locating the Voronoi cell containing the query point.
NDC searches for the target as follows, from the starting query Qs,
k points are randomly retrieved.
Then the Voronoi region VRi is initially set to the minimum bounding box of S.
15. Algorithm
Instead of using a query point and its neighboring points to construct a Voronoi diagram,
GDC uses the query point and k points randomly sampled from V Ri.