International Journal on Document Analysis and Recognition最新文献

Digitizing historical tabular records is essential for preserving and analyzing valuable data across various fields, but it presents challenges due to complex layouts, mixed text types, and degraded document quality. This paper introduces a comprehensive framework to address these issues through three key contributions. First, it presents UoS_Data_Rescue, a novel dataset of 1,113 historical logbooks with over 594,000 annotated text cells, designed to handle the complexities of handwritten entries, aging artifacts, and intricate layouts. Second, it proposes a novel context-aware text extraction approach (TrOCR-ctx) to reduce cascading errors during table digitization. Third, it proposes an enhanced end-to-end OCR pipeline that integrates TrOCR-ctx with ByT5, combining OCR and post-OCR correction in a unified training framework. This framework enables the system to produce both the raw OCR output and a corrected version in a single pass, improving recognition accuracy, particularly for multilingual and degraded text, within complex table digitization tasks. The model achieves superior performance with a 0.049 word error rate and a 0.035 character error rate, outperforming existing methods by up to 41% in OCR tasks and 10.74% in table reconstruction tasks. This framework offers a robust solution for large-scale digitization of tabular documents, extending its applications beyond climate records to other domains requiring structured document preservation. The dataset and implementation are available as open-source resources.

The redaction of sensitive information in documents is common practice in specific types of organizations. This happens for example in court proceedings or in documents released under the Freedom of Information Act (FOIA). The ability to automatically detect when information has been redacted has several practical applications, such as the gathering of statistics on the amount of redaction present in documents, enabling a critical view on redaction practices. It can also be used to further investigate redactions, and whether or not the used techniques provide sufficient anonymization. The task is particularly challenging because of the large variety of redaction methods and techniques, from software for automatic redaction to manual redactions by pen. Any detection system must be robust to a large variety of inputs, as it will be run on many documents that might not even contain redactions. In this study, we evaluate two neural methods for the task, namely a Mask R-CNN model and a Mask2Former model, and compare them to a rule-based model based on optical character recognition and morphological operations. The best performing, the Mask R-CNN model, has a recall of .94 with a precision of .96 over a challenging data set containing several redaction types. Adding many pages without redaction barely lowers this score (precision drops to .90, recall drops to .92). The Mask2Former model is most robust to inputs without redactions, producing the least false positives of all models.

Human personality is a blend of different traits and virtues. It’s modeling is challenging due to its inherent complexity. There are multitudinous cues to predict personality and handwriting is one of them. This is because it is distinctive to a large extent and varies at the individual level. The allied field of science which deals with the analysis of handwriting for understanding personality is known as Graphology. Researchers have discovered disparate features of handwriting that can reveal the personality traits of an individual. Several attempts have been made to model personality from handwriting in different languages but significant advancement is required for commercialization. In this paper, we present the reported aspects of handwriting, techniques for processing handwritten documents and evaluation measures for personality identification to draw a horizon and aid in further advancement of research in this field.

Data availability is a big concern in the field of document analysis, especially when working on tasks that require a high degree of precision when it comes to the definition of the ground truths on which to train deep learning models. A notable example is represented by the task of document layout analysis in handwritten documents, which requires pixel-precise segmentation maps to highlight the different layout components of each document page. These segmentation maps are typically very time-consuming and require a high degree of domain knowledge to be defined, as they are intrinsically characterized by the content of the text. For this reason in the present work, we explore the effects of different initialization strategies for deep learning models employed for this type of task by relying on both in-domain and cross-domain datasets for their pre-training. To test the employed models we use two publicly available datasets with heterogeneous characteristics both regarding their structure as well as the languages of the contained documents. We show how a combination of cross-domain and in-domain transfer learning approaches leads to the best overall performance of the models, as well as speeding up their convergence process.

The field of computer vision has seen significant transformation with the emergence and advancement of deep learning models. Deep learning waves have a significant impact on scene text detection, a vital and active area in computer vision. Numerous scientific, industrial, and academic procedures make use of text analysis. Natural scene text detection is more difficult than document image text detection owing to variations in font, size, style, brightness, etc. The National Institute of Technology Jalandhar-Text Detection dataset (NITJ-TD) is a new dataset that we have put forward in this study for various text analysis tasks including text detection, text segmentation, script identification, text recognition, etc. a deep learning model that seeks to identify the text’s location within the image,which are gathered in an unrestricted setting. The system consists of an NMS to choose the best match and prevent repeated predictions, and a modified EAST to pinpoint the exact ROI in the image. To improve the model’s performance, an enhancement module is added to the fundamental Efficient and Accurate Scene Text detector (EAST). The suggested approach is contrasted in terms of text word detection in the image. Several pre-trained models are used to assign the text word to various intersections over Union (IoU) values. We made use of our NITJ-TD dataset, which is made up of 1500 photos that were gathered from various North Indian sites. Punjabi, English, and Hindi scripts can be seen on the images. We also examined the outcomes of the ICDAR-2013 benchmark dataset. On both the suggested dataset and the benchmarked dataset, our approach performed better.

Construction drawings are frequently stored in undigitised formats and consequently, their analysis requires substantial manual effort. This is true for many crucial tasks, including material takeoff where the purpose is to obtain a list of the equipment and respective amounts required for a project. Engineering drawing digitisation has recently attracted increased attention, however construction drawings have received considerably less interest compared to other types. To address these issues, this paper presents a novel framework for the automatic processing of construction drawings. Extensive experiments were performed using two state-of-the-art deep learning models for object detection in challenging high-resolution drawings sourced from industry. The results show a significant reduction in the time required for drawing analysis. Promising performance was achieved for symbol detection across various classes, with a mean average precision of 79% for the YOLO-based method and 83% for the Faster R-CNN-based method. This framework enables the digital transformation of construction drawings, improving tasks such as material takeoff and many others.

Text line segmentation (TLS) is an essential step of the end-to-end document analysis systems. The main purpose of this step is to extract the individual text lines of any handwritten documents with high accuracy. Handwritten and historical documents mostly contain touching and overlapping characters, heavy diacritics, footnotes and side notes added over the years. In this work, we present a new TLS method based on generative adversarial networks (GAN). TLS problem is tackled as an image-to-image translation problem and the GAN model was trained to learn the spatial information between the individual text lines and their corresponding masks including the text lines. To evaluate the segmentation performance of the proposed GAN model, two challenging datasets, VML-AHTE and VML-MOC, were used. According to the qualitative and quantitative results, the proposed GAN model achieved the best segmentation accuracy on the VML-MOC dataset and showed competitive performance on the VML-AHTE dataset.

Classification of Palm leaf images into various quality categories is an important step towards the digitization of these heritage documents. Manual inspection and categorization is not only laborious, time-consuming and costly but also subject to inspector’s biases and errors. This study aims to automate the classification of palm leaf document images into three different visual quality categories. A comparative analysis between various structural and statistical features and classifiers against deep neural networks is performed. VGG16, VGG19 and ResNet152v2 architectures along with a custom CNN model are used, while Discrete Cosine Transform (DCT), Grey Level Co-occurrence Matrix (GLCM), Tamura, and Histogram of Gradient (HOG) are chosen from the traditional methods. Based on these extracted features, various classifiers, namely, k-Nearest Neighbors (k-NN), multi-layer perceptron (MLP), Support Vector Machines (SVM), Decision Tree (DT) and Logistic Regression (LR) are trained and evaluated. Accuracy, precision, recall, and F1 scores are used as performance metrics for the evaluation of various algorithms. Results demonstrate that CNN embeddings and DCT features have emerged as superior features. Based on these findings, we integrated DCT with a Vision Transformer (ViT) for the document classification task. The result illustrates that this incorporation of DCT with ViT outperforms all other methods with 96% train F1 score and a test F1 score of 90%.

Table detection, a pivotal task in document analysis, aims to precisely recognize and locate tables within document images. Although deep learning has shown remarkable progress in this realm, it typically requires an extensive dataset of labeled data for proficient training. Current CNN-based semi-supervised table detection approaches use the anchor generation process and non-maximum suppression in their detection process, limiting training efficiency. Meanwhile, transformer-based semi-supervised techniques adopted a one-to-one match strategy that provides noisy pseudo-labels, limiting overall efficiency. This study presents an innovative transformer-based semi-supervised table detector. It improves the quality of pseudo-labels through a novel matching strategy combining one-to-one and one-to-many assignment techniques. This approach significantly enhances training efficiency during the early stages, ensuring superior pseudo-labels for further training. Our semi-supervised approach is comprehensively evaluated on benchmark datasets, including PubLayNet, ICADR-19, and TableBank. It achieves new state-of-the-art results, with a mAP of 95.7% and 97.9% on TableBank (word) and PubLaynet with 30% label data, marking a 7.4 and 7.6 point improvement over previous semi-supervised table detection approach, respectively. The results clearly show the superiority of our semi-supervised approach, surpassing all existing state-of-the-art methods by substantial margins. This research represents a significant advancement in semi-supervised table detection methods, offering a more efficient and accurate solution for practical document analysis tasks.

Most molecular diagram parsers recover chemical structure from raster images (e.g., PNGs). However, many PDFs include commands giving explicit locations and shapes for characters, lines, and polygons. We present a new parser that uses these born-digital PDF primitives as input. The parsing model is fast and accurate, and does not require GPUs, Optical Character Recognition (OCR), or vectorization. We use the parser to annotate raster images and then train a new multi-task neural network for recognizing molecules in raster images. We evaluate our parsers using SMILES and standard benchmarks, along with a novel evaluation protocol comparing molecular graphs directly that supports automatic error compilation and reveals errors missed by SMILES-based evaluation. On the synthetic USPTO benchmark, our born-digital parser obtains a recognition rate of 98.4% (1% higher than previous models) and our relatively simple neural parser for raster images obtains a rate of 85% using less training data than existing neural approaches (thousands vs. millions of molecules).