Incremental dimension reduction for efficient and accurate visual anomaly detection

2026-02-27

AI Summary

This work addresses the computational and memory efficiency bottlenecks faced by existing visual anomaly detection methods when processing large-scale image data due to high-dimensional features. To overcome this challenge, the paper proposes an incremental dimensionality reduction framework that, for the first time, introduces Incremental Truncated Singular Value Decomposition (Incremental Truncated SVD) into this domain. By processing deep features in batches and dynamically updating a low-dimensional subspace, the approach achieves both memory efficiency and global feature consistency. Each batch’s representation is mapped back to a unified global feature space, enabling efficient and accurate anomaly detection. Experimental results demonstrate that the proposed method significantly accelerates the training of state-of-the-art detection algorithms while maintaining detection accuracy comparable to that of original high-dimensional approaches.

Technology Category

Machine Learning: Dimensionality Reduction/Feature SelectionComputer Vision: Learning & Optimization for CVData Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphsSearch and Retrieval-Augmented AI: Efficiency and scalability of Web search enginesWeb Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

While nowadays visual anomaly detection algorithms use deep neural networks to extract salient features from images, the high dimensionality of extracted features makes it difficult to apply those algorithms to large data with 1000s of images. To address this issue, we present an incremental dimension reduction algorithm to reduce the extracted features. While our algorithm essentially computes truncated singular value decomposition of these features, other than processing all vectors at once, our algorithm groups the vectors into batches. At each batch, our algorithm updates the truncated singular values and vectors that represent all visited vectors, and reduces each batch by its own singular values and vectors so they can be stored in the memory with low overhead. After processing all batches, we re-transform these batch-wise singular vectors to the space spanned by the singular vectors of all features. We show that our algorithm can accelerate the training of state-of-the-art anomaly detection algorithm with close accuracy.

Problem

• visual anomaly detection
• high dimensionality
• large-scale data
• feature extraction
• computational efficiency

Innovation

• incremental dimension reduction
• truncated SVD
• visual anomaly detection
• batch processing
• feature compression