Bag of Bags: Adaptive Visual Vocabularies for Genizah Join Image Retrieval

2026-04-09

AI Summary

This work addresses the problem of reconnecting fragmented manuscripts from the Cairo Genizah by proposing a "Bag-of-Bags" (BoB) representation. Instead of relying on a global visual vocabulary, BoB constructs an adaptive local visual codebook for each image. Binary patch features are extracted via a sparse convolutional autoencoder, and set representations are formed by combining connected component analysis with image-level k-means clustering. Fragment matching is performed based on distances between these sets. The authors further introduce BoB-OT, a variant that incorporates cluster-size weighting and provides theoretical approximation guarantees with respect to full-component optimal transport discrepancy. Evaluated on real-world data, their two-stage retrieval pipeline—comprising an initial Bag-of-Words filtering followed by BoB-OT reranking—achieves a Hit@1 of 0.78 and MRR of 0.84, outperforming the strongest BoW baseline by 6.1% in top-1 accuracy while maintaining computational efficiency.

Technology Category

• Computer Vision: Image and Video Retrieval
• Search and Optimization: Distributed Search
• Machine Learning: Representation Learning

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

A join is a set of manuscript fragments identified as originally emanating from the same manuscript. We study manuscript join retrieval: Given a query image of a fragment, retrieve other fragments originating from the same physical manuscript. We propose Bag of Bags (BoB), an image-level representation that replaces the global-level visual codebook of classical Bag of Words (BoW) with a fragment-specific vocabulary of local visual words. Our pipeline trains a sparse convolutional autoencoder on binarized fragment patches, encodes connected components from each page, clusters the resulting embeddings with per image $k$-means, and compares images using set to set distances between their local vocabularies. Evaluated on fragments from the Cairo Genizah, the best BoB variant (viz.\@ Chamfer) achieves Hit@1 of 0.78 and MRR of 0.84, compared to 0.74 and 0.80, respectively, for the strongest BoW baseline (BoW-RawPatches-$χ^2$), a 6.1\% relative improvement in top-1 accuracy. We furthermore study a mass-weighted BoB-OT variant that incorporates cluster population into prototype matching and present a formal approximation guarantee bounding its deviation from full component-level optimal transport. A two-stage pipeline using a BoW shortlist followed by BoB-OT reranking provides a practical compromise between retrieval strength and computational cost, supporting applicability to larger manuscript collections.

Problem

• manuscript join retrieval
• fragment matching
• visual vocabulary
• image retrieval
• Cairo Genizah

Innovation

• Bag of Bags
• fragment-specific vocabulary
• sparse convolutional autoencoder
• set-to-set distance
• optimal transport