End-to-end OCR increasingly relies on autoregressive sequence models, where the quadratic cost of Transformer attention limits efficient transcription of long, paragraph-level text. State-Space Models (SSMs) such as Mamba offer linear-time decoding and have recently been shown to match Transformer accuracy on printed historical lines, but their behavior as sequences grow from short lines to full paragraphs, and their generalization to handwriting, remain poorly understood. We study how a Mamba-based OCR recognizer scales from lines to paragraphs. We first conduct a systematic exploration of its four core hyperparameters (decoder depth, state dimension, expansion factor, and connector depth) on synthetic paragraphs from 100 to 1,000 characters, identifying the recurrent state dimension and the expansion factor as the dominant levers for long-sequence accuracy. We then compare the recognizer against a Transformer baseline trained under an identical protocol. On clean synthetic paragraphs, both models stay below 1

End-to-end OCR for historical newspapers remains challenging, as models must handle long text sequences, degraded print quality, and complex layouts. While Transformer-based recognizers dominate current research, their quadratic complexity limits efficient paragraph-level transcription and large-scale deployment. We investigate linear-time State-Space Models (SSMs), specifically Mamba, as a scalable alternative to Transformer-based sequence modeling for OCR. We present to our knowledge, the first OCR architecture based on SSMs, combining a CNN visual encoder with bi-directional and autoregressive Mamba sequence modeling, and conduct a large-scale benchmark comparing SSMs with Transformer- and BiLSTM-based recognizers. Multiple decoding strategies (CTC, autoregressive, and non-autoregressive) are evaluated under identical training conditions alongside strong neural baselines (VAN, DAN, DANIEL) and widely used off-the-shelf OCR engines (PERO-OCR, Tesseract OCR, TrOCR, Gemini). Experiments on historical newspapers from the Bibliothèque nationale du Luxembourg, with newly released >99% verified gold-standard annotations, and cross-dataset tests on Fraktur and Antiqua lines, show that all neural models achieve low error rates ( ∼ 2 × vs 2.30 × growth at 1000 chars), reaching 6.07 × faster. We release code, trained models, and standardized evaluation protocols to enable reproducible research and guide practitioners in large-scale cultural heritage OCR available at https://github.com/MarcoPerson/ssm-ocr-benchmark .

Understanding newspaper images remains a challenging task due to their complex, nested hierarchical structures and dense, heterogeneous layouts. In this paper, we explore two complementary approaches for newspaper structure understanding. First, we present a modular bottom-up pipeline that combines state-of-the-art open-source models: YOLO for layout detection, LayoutReader for reading order prediction, and a custom algorithm for article segmentation. This approach leverages existing robust components while maintaining flexibility and interpretability. Second, we introduce Tiramisu (Tiered Transformers for Hierarchical Structure Understanding), a novel end-to-end transformer-based architecture that explicitly models document hierarchy through an iterative tiered process. Tiramisu performs section and article separation, block localization, semantic categorization, and reading order prediction using highly parallelized attention mechanisms. Finally, we release Finlam La Liberté, a new dataset designed specifically for evaluating hierarchical information retrieval in historical newspapers. Experimental results demonstrate the effectiveness of both approaches in reconstructing complex newspaper hierarchies, with comparative analysis highlighting their respective strengths for scalable document digitization. The Tiramisu training code, including the synthetic newspaper generator, is available at https://git.litislab.fr/tiramisu/tiramisu-newspaper-articles-extractor.

This paper describes a new segmentation-free method for retrieving images and spotting patterns in historical document image collections. The proposed method needs no training on the target domain, characterizing a problem-independent approach. For this purpose, the query and the document image represented by feature maps extracted using intermediate layers of a pre-trained Fully Convolutional Network are submitted to a cross-correlation process. The produced similarity heatmap is used to locate the query occurrences on the document page. A robust experimental protocol using three datasets shows promising results on image retrieval and pattern spotting. The experiments conducted on the public DocExplore dataset demonstrated that the proposed method could improve the mAP by 69.1

The goal of this paper is to propose an unsupervised learning-based framework in order to deal with any kind of one-shot object detection scenario, focusing on the tasks of sub-image retrieval and pattern spotting in historical document images. Taking in an arbitrary object/pattern query from users, the proposed framework should be able to retrieve images containing it, as well as localising each occurrence within the images. A major difficulty is the lack of any training data. Three contributions are thus presented: (1) a novel model architecture dubbed OS-DETR, capable of adapting to various tasks by simply swapping training data, (2) a completely unsupervised synthetic data generation process, easily applicable to many data-limited domains, and (3) a set of training strategies catered to boost the model's generalisation capabilities. The result is a framework that yields a strong baseline for learning-based approaches applied to sub-image retrieval and pattern spotting.

This paper describes a method to perform the tasks of Image Retrieval and Pattern Spotting in a collection of historical documents, in a zero-shot manner, i.e. without prior knowledge or any training on the pattern to be searched. The proposed method measures the similarity between images using representation schemes based on feature maps provided by intermediate layers of a CNN. Moreover, to improve the time response and reduce the storage need, we propose to binarize these image representations. Experimental results obtained on the DocExplore dataset show that the proposed method improves the mAP on this dataset by 38.46% for Image Retrieval and 134.6% for Pattern Spotting compared to the state-of-the-art methods. The proposed binarization strategy provides a reduction of memory usage by a factor of 16 with a slight decrease of less than one percentage point in the mAP for both tasks and a reduction of 13.3% in search time when using the binary representation.

In this paper, we propose a hybrid system for contextual and semantic indexing of Arabic documents, bringing an improvement to classical models based on n-grams and the Okapi model. This new approach takes into account the concept of the semantic vicinity of terms. We proceed in fact by the calculation of similarity between words using an hybridization of NGRAMs-OKAPI statistical measures and a kernel function in order to identify relevant descriptors. Terminological resources such as graphs and semantic dictionaries are integrated into the system to improve the indexing and the classification processes.

Historical newspaper image analysis is a challenging task due to the complex layout of newspapers and its variability among collections. While traditional approaches are rule-based methods with many successive steps, recent works show that deep learning approaches can be successfully used to provide a pixel labeling of the various fields occurring in a page. This allows the automatic extraction of the document structure and accessing the different semantic entities. Recent improvements proposed to strengthen convolutional neural network capacities such as gated mechanism may also apply well to to task at end. In this respect, we propose a fully convolutional neural network architecture (FCN) that outputs a pixel-labeling of the various semantic entities that occur in historical newspaper images. Our model is based on a novel Multi-Scale Gated Block architecture (MSGB), made of dense connections and gating mechanisms that handle a multi-scale analysis of the input image with self-attention. Evaluations conducted on 4 historical newspaper datasets including up to 11 semantic classes show that our proposition outperforms standard FCN architectures. (c) 2020 Elsevier B.V. All rights reserved.

Pattern spotting consists of locating different instances of a given object (i.e. an image query) in a collection of historical document images. These patterns may vary in shape, size, color, context and even style because they are hand-drawn, which makes pattern spotting a difficult task. To tackle this problem, we propose a Convolutional Neural Network (CNN) approach based on Feature Pyramid Networks (FPN) as the feature extractor of our system. Using FPN allows to extract descriptors of local regions of the documents to be indexed and queries, at multiple scales with just a single forward pass. Experiments conducted on DocExplore dataset show that the proposed system improves mAP by 73% (from 0.157 to 0.272) in pattern localization compared with state-of-the-art results, even when the feature extractor is not trained with domain-specific data. Memory requirement and computation time are also decreased since the descriptor dimension used for distance computation is reduced by a factor of 16. (c) 2020 Published by Elsevier B.V.

The French National Committee for Scientific Diving (CNPS) is an inter-agency committee on scientific diving activities. French marine universities, research institutes (CNRS-INEE/INSU/INSHS, DRASSM, EPHE, IFREMER, INRA, INSERM, IPEV, IRD, IRSTEA) and the National Museum (MNHN) are involved or represented. The CNPS, created in 1999, is the scientific diving representative of the Ministry of Higher Education and Research to the Ministry of Labor. It includes personalities chosen for their expertise in the field of professional use of diving (scientific and technical diving, training), the law of the sea, the hyperbaric medicine and the European scientific diving panel (ESDP). The aims of this group of experts are: a) to observe the evolution of the professional scientific diving practices worldwide, b) to define common frameworks and promote the best practices, c) to provide high-level expertise and advices to Ministries of Labor and Research, d) to monitor the innovations in scientific diving, technology, field practices and hyperbaric medicine, e) to share informations and feedback with the French scientific divers community, f) to participate in the training of scientific divers, and g) to represent France at the ESDP. The CNPS has been fully involved since 2009 in the formulation of new laws and regulations dedicated to professional diving with the Ministry of Labor. The French scientific diving community is very diversified, for historical reasons, due to its 18 000 km long coastline and a strong presence overseas, through the involvement of researchers in global science issues and international collaborations. This poster presents the aims, activities and key results obtained by the CNPS. It also sets out some perspectives, especially actions in favour of a convergence of European practice patterns and standards.

Pattern spotting consists of searching in a collection of historical document images for occurrences of a graphical object using an image query. Contrary to object detection, no prior information nor predefined class is given about the query so training a model of the object is not feasible. In this paper, a convolutional neural network approach is proposed to tackle this problem. We use RetinaNet as a feature extractor to obtain multiscale embeddings of the regions of the documents and also for the queries. Experiments conducted on the DocExplore dataset show that our proposal is better at locating patterns and requires less storage for indexing images than the state-of-the-art system, but fails at retrieving multiple pages containing instances of the query.

With advances in technology, a large part of our cultural heritage is becoming digitally available. In particular, in the field of historical document image analysis, there is now a growing need for indexing and data mining tools, thus allowing us to spot and retrieve the occurrences of an object of interest, called a pattern, in a large database of document images. Patterns may present some variability in terms of color, shape, or context, making the spotting of patterns a challenging task. Pattern spotting is a relatively new field of research, still hampered by the lack of available annotated resources. We present a new publicly available dataset named DocExplore dedicated to spotting patterns in historical document images. The dataset contains 1500 images and 1464 queries, and allows the evaluation of two tasks: image retrieval and pattern localization. A standardized benchmark protocol along with ad hoc metrics is provided for a fair comparison of the submitted approaches. We also provide some first results obtained with our baseline system on this new dataset, which show that there is room for improvement and that should encourage researchers of the document image analysis community to design new systems and submit improved results. (C) 2016 SPIE and IS&T

Template matching is a classical and essential step in many pattern recognition, object detection or video tracking systems. This paper aims at integrating and evaluating different template matching methods in the context of pattern spotting in historical document images - i.e. the search for occurrences of a given visual pattern in document images. Given a query image, our pattern spotting system first computes the similarity score between the query signature and the signatures of a few regions provided by a region proposal algorithm. The top ranked regions are then selected for further processing. Template Matching is then applied in the neighborhood of the selected regions to precisely locate and rank the candidate windows that maximize the matching score. This paper builds upon popular template matching approaches and provides a unified testing framework for historical document image pattern spotting. The experimentation offers an insight on how to choose the most promising techniques for historical document images. This paper also proposes an improvement over these standard template matching approaches to significantly increase the overall performance.

Pattern spotting consists in searching in a document image for the occurrences of a queried graphical object. The main challenge in pattern spotting is that the query image is generally small and the occurrences may be located at any random places in the image. Rather than exhaustively indexing all possible subwindows extracted from the document images, the common way is to rely on a segmentation or a document layout analysis to limit the search space. However, there is no segmentation nor document layout analysis technique reliable enough for historical document images. Region proposal, a technique used to generate a set of regions potentially containing an object, has contributed to many state of the art object detection systems recently. Although it is initially proposed for object detection, we will show that region proposal also offers promising results for document images, particularly in the case of pattern spotting. In this paper, we aim at investigating the use of region proposal to produce high quality subwindows to replace the usual document layout analysis step and the blind sliding windowing step. From experiments conducted on the DocExplore dataset, we show that region proposal generates a comparable number of subwindows while helping the system to achieve significant better results than the system built with commonly used layout analysis techniques.

Information retrieval in historical documents has long consisted in spotting words. In this paper, we focus on graphical pattern spotting. Contrary to object detection and classification, where models of the object of interest may be trained, pattern spotting does not rely on any prior information regarding the query, nor predefined class of graphical objects. An offline sliding window approach may be suitable, provided that the challenge raised by high computational and storage costs is handled. We propose an unsupervised, segmentation-free approach that takes advantage of recent developments in computer vision to overcome these issues. We also investigate the use of new, compact descriptors for the data, namely the vectors of locally aggregated descriptors (VLAD) and Fisher Vectors, instead of the usual bag-of-visual-words approach. Results obtained on medieval manuscripts from the DocExplore project show that our approach achieves better retrieval results, with a better efficiency in terms of time/memory, compared to standard approaches. Experimentations show that VLAD and Fisher Vectors can be fruitfully used in the future for the description of historical documents. Additionally, we show that our system can be easily turned into a word spotting system with slight adaptation, and that it achieves results comparable to those recently published in ICDAR 2015 keyword spotting challenge.

PIVAJ is a platform for archived digitized newspaper emphasizing articles: extracting them from digitized documents by automated page layout analysis, OCRing them, indexing their text transcription to allow users to search for content. Crowdsourcing is used to improve the quality of the indexing, by correcting the transcription and by tagging articles with keywords. The platform has been used to give Web access to 550 000 articles generated from a digitized local newspaper. Current developments include further improvements to its OCR as well as graphical interfaces for the management of the platform.

Pattern spotting consists of retrieving the most similar graphical patterns from a collection of document images. Inspired by the recent advances in computer vision and word spotting techniques, we propose in this paper an unsupervised, segmentation-free pattern spotting system. Overall, the system includes a powerful patch-based framework, the bag of visual word model with an offline sliding window mechanism to avoid heavy computational burden during the retrieval process. Our system takes advantage of the most recent powerful compression and distance approximation techniques (product quantization and asymmetric distance computation) to efficiently index the great number of sub-windows produced by sliding windows and allows to retrieve small sized queries in a large indexed corpus.

This paper introduces a new distance function for comparing images in the context of content-based image retrieval. Given a query and a large dataset to be searched, the system has to provide the user – as efficiently as possible – with a list of images ranked according to their distance to the query. Because of computational issues, traditional image search systems are generally based on conventional distance function such as the Euclidian distance or the dot product, avoiding the use of any training data nor expensive online metric learning algorithms. The drawback is that, in this case, the system can hardly cope with the variability of image contents. This paper proposes a simple yet efficient zero-shot learning algorithm that can learn a query-adapted distance function from a single image (the query) or from a few images (e.g. some user-selected images in a relevance feedback iteration), hence improving the quality of the retrieved images. This allows our system to work with any object categories without requiring any training data, and is hence more applicable in real world use cases. More interestingly, our system can learn the metric on the fly, at almost no cost, and the cost of the ranking function is as low as the dot product distance. By allowing the system to learn to rank the images, significantly and consistently improved results (over the conventional approaches) have been observed on the Oxford5k, Paris6k and Holiday1k datasets.