[Paper Review] PCA-Based Relevance Feedback in Document Image Retrieval

Reza Tavoli, Fariborz Mahmoudi|arXiv (Cornell University)|Sep 11, 2012

Image Retrieval and Classification Techniques15 references3 citations

TL;DR

This paper proposes a PCA-based relevance feedback framework for document image retrieval (DIR) that dynamically updates feature subspaces using Principal Component Analysis to improve retrieval accuracy. By incorporating user feedback to refine both positive and negative examples and projecting features into optimized subspaces, the method achieves superior performance over conventional DIR systems, particularly in reducing dimensionality and enhancing relevance precision.

ABSTRACT

Research has been devoted in the past few years to relevance feedback as an effective solution to improve performance of information retrieval systems. Relevance feedback refers to an interactive process that helps to improve the retrieval performance. In this paper we propose the use of relevance feedback to improve document image retrieval System (DIRS) performance. This paper compares a variety of strategies for positive and negative feedback. In addition, feature subspace is extracted and updated during the feedback process using a Principal Component Analysis (PCA) technique and based on user's feedback. That is, in addition to reducing the dimensionality of feature spaces, a proper subspace for each type of features is obtained in the feedback process to further improve the retrieval accuracy. Experiments show that using relevance Feedback in DIR achieves better performance than common DIR.

Motivation & Objective

• To enhance document image retrieval (DIR) performance using relevance feedback.
• To address the challenge of feature space redundancy and irrelevance in DIR by dynamically adapting feature subspaces.
• To integrate user feedback into a PCA-based framework that refines feature representation over iterations.
• To compare various feedback strategies (positive/negative) in improving retrieval accuracy.
• To demonstrate that subspace adaptation via PCA leads to better retrieval performance than standard DIR approaches.

Proposed method

• The system employs relevance feedback where users label relevant and non-relevant documents to guide retrieval refinement.
• Principal Component Analysis (PCA) is applied to extract and update a reduced, optimized feature subspace based on user feedback.
• Feature subspaces are retrained iteratively using feedback data to focus on discriminative dimensions, improving representation quality.
• The method supports multiple feedback strategies, including balanced and unbalanced positive/negative feedback, to assess robustness.
• Dimensionality reduction via PCA reduces noise and computational load while preserving discriminative information.
• The framework integrates feedback into the retrieval pipeline by reweighting or reprojecting features into the learned subspace.

Experimental results

Research questions

• RQ1 How does relevance feedback improve document image retrieval performance when combined with PCA-based feature subspace adaptation?
• RQ2 Which feedback strategy—positive-only, negative-only, or combined—yields the highest retrieval accuracy?
• RQ3 To what extent does PCA-based subspace learning enhance retrieval precision compared to standard DIR methods?
• RQ4 Can iterative feedback and dynamic subspace updates lead to consistent performance gains across different document image datasets?
• RQ5 How does the dimensionality reduction via PCA affect the robustness and efficiency of the retrieval system?

Key findings

• The proposed PCA-based relevance feedback method significantly improves retrieval accuracy compared to baseline DIR systems.
• Combined positive and negative feedback with PCA subspace adaptation yields the highest performance gain, outperforming single-feedback strategies.
• Dimensionality reduction via PCA reduces feature space noise and enhances retrieval precision by focusing on discriminative components.
• Iterative feedback with dynamic subspace updates leads to consistent performance improvements across multiple retrieval iterations.
• The method achieves better precision and recall than conventional DIR systems, particularly in complex document image collections.
• The use of PCA enables efficient and effective feature space transformation that adapts to user relevance judgments over time.