Fast Query Point Movement Techniques with Relevance Feedback for Content-Based Image Retrieval

Abstract

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, 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 relevance feedback. Extensive experiments in simulated and realistic environments show that our approach significantly reduces the number of required iterations and improves overall retrieval performance. The experimental results also confirm that our approach can always retrieve intended targets even with poor selection of initial query points.

Medical image retrieval: Multiple regression models for user's search target

“…In the beginning the system has been initially run by a query images without any feedback and so the average precision rate is 30% for all methods, i.e., Logistic Regression (LR), SVM classification (SVM) [8] and Query Point Movement (QPM) [9]. After sev-…”

Fast Query Point Movement Techniques for Large CBIR Systems

“…The two images (i.e., an ancient building and race cars) took, on average, more iterations than the others to retrieve, mainly because many similar images exist in the collection. Even so, only 6 iterations on average were needed to locate them, while 7 iterations were needed without our index structure [21]. The results illustrate that our index structure can help reduce the number of iterations.…”

“…Target search may involve four typical types of queries: sampling queries [8], [10], [12], [17], [18], [20], [21], [26], constrained sampling queries [21], k-NN queries [18], [26] and constrained k-NN queries [21] as discussed in Section III. Among all the aforementioned techniques, only Chakrabarti et al discussed how to efficiently evaluated k-NN queries in the Query Expansion model [8] for category search.…”

“…In practice, however, this cannot be easily achieved by human evaluators, unless the most relevant images are distinctly stood out. To evaluate our methods' performance in realistic environments, we have extended the previous prototype [21] (based …”

“…Our extensive experimental results confirm our complexity analysis, and show the advantage of the proposed techniques in both simulated and realistic environments. A preliminary version of this study was presented in [21]. In the current paper, we extend the original technique to consider inaccurate user relevance feedback.…”