International Conference On Document Analysis And Recognition Research Articles

Advancements and Challenges in Handwritten Text Recognition: A Comprehensive Survey.

Handwritten Text Recognition (HTR) is essential for digitizing historical documents in different kinds of archives. In this study, we introduce a hybrid form archive written in French: the Belfort civil registers of births. The digitization of these historical documents is challenging due to their unique characteristics such as writing style variations, overlapped characters and words, and marginal annotations. The objective of this survey paper is to summarize research on handwritten text documents and provide research directions toward effectively transcribing this French dataset. To achieve this goal, we presented a brief survey of several modern and historical HTR offline systems of different international languages, and the top state-of-the-art contributions reported of the French language specifically. The survey classifies the HTR systems based on techniques employed, datasets used, publication years, and the level of recognition. Furthermore, an analysis of the systems' accuracies is presented, highlighting the best-performing approach. We have also showcased the performance of some HTR commercial systems. In addition, this paper presents a summarization of the HTR datasets that publicly available, especially those identified as benchmark datasets in the International Conference on Document Analysis and Recognition (ICDAR) and the International Conference on Frontiers in Handwriting Recognition (ICFHR) competitions. This paper, therefore, presents updated state-of-the-art research in HTR and highlights new directions in the research field.

Local binary pattern and its derivatives to handwriting-based gender classification

Several studies by psychologists and computer scientists have verified the link between handwriting and writer gender. The texture of the writing image is a major indicator of whether it is male or female writing. This paper conducts a comparison analysis to examine the effectiveness of various local binary patterns (LBPs) techniques in detecting gender from scanned images of handwriting. We study different LBP variants, including complete local binary pattern (CLBP), local ternary pattern (LTP), local configuration pattern (LCP), rotated local binary pattern (RLBP), local binary pattern variance (LBPV), and multi-scale local binary pattern (MLBP), as features for representing handwriting images. A support vector machine (SVM) is trained using features from male and female writing. The method achieves encouraging classification rates of 76.68 when tested on subsets of the Qatar University writer identification (QUWI) dataset containing English and Arabic writing samples when using the experimental protocols of the International Conference on Document Analysis and Recognition (ICDAR) 2013 gender classification competitions.

Feature fusion-based text information mining method for natural scenes

Abstract As a crucial medium of information dissemination, text holds a pivotal role in a multitude of applications. However, text detection in complex and unstructured environments presents significant challenges, such as the presence of cluttered backgrounds, variations in appearance, and uneven lighting conditions. To address this issue, this study proposes a text detection framework that leverages multistage edge detection and contextual information. This framework deviates from traditional approaches by incorporating four primary processing steps, including text visual saliency region detection to accentuate the text regions and diminish background interference, multistage edge detection to enhance the conventional stroke width transform results, a texture-based and connected components-based integration to accurately distinguish text from the background, and a context fusion step to recover missing text regions and improve the recall of text detection. The proposed method was evaluated on two widely used benchmark datasets, i.e., the international conference on document analysis and recognition (ICDAR) 2005 dataset and the ICDAR 2011 dataset, and the results indicate the advancedness of the method.

Identification of Proper and Improper Signatures Using Graph Theory Techniques

Objectives: To propose an automatic signature identification for off-line signature utilising graph theory approaches. Methods: Scanned signatures (Kaggle, https://www.kaggle.com/divyanshrai/handwritten-signatures/data) are collected for off-line signature data. The method follows pre-processing, vertex point extraction by midpoint traverse method, features extraction using edge, average edge and average edge D-distance and Support Vector Machine (SVM) to classify and predict the true label for the genuine and forged signatures. False Acceptance Ratio (FAR) and False Rejection Ratio (FRR) give the accuracy of the proposed methods. This off-line signature verification method is compared with the deep learning techniques existing in the literature. Findings: Support Vector Machine (SVM) used for classification and results on standard signature datasets like ICDAR (International Conference on Document Analysis and Recognition). The results demonstrate how the proposed strategy outperforms the state-of-the-art already available. Novelty: The proposed approach use the edge distance, average edge distance, and average edge D-distance inbuilt graph structures to extract the feature points. Keywords: Signature images; grid approach; bipartite graph; complete bipartite graph; mid point traverse method

A Fast High Precision Skew Angle Estimation of Digitized Documents

In this paper, we treated the problem of automatic skew angle estimation of scanned documents. The skew of document occurs very often, due to incorrect positioning of the documents or a manipulation error during scanning. This has negative consequences on the steps of automatic analysis and recognition of text. It is therefore essential to verify, before proceeding to these steps, the presence of skew on the document to be processed and to correct it. The difficulty of this verification is associated to the presence of graphic zones, sometimes dominant, that have a considerable impact on the accuracy of the text skew angle estimation. We also noted the importance of preprocessing to improve the accuracy and the calculation cost of skew estimation approaches. These two elements have been taken into consideration in our design and development of a new approach of skew angle estimation and correction. Our approach is based on local binarization followed by horizontal smoothing by the Run Length Smoothing Algorithm (RLSA) method, detection of horizontal contours and the Hierarchical Hough Transform (HHT). The algorithms involved in our approach have been chosen to guarantee a skew estimation: accurate, fast and robust, especially to graphic dominance and real time application. The experimental tests show the effectiveness of our approach on a representative database of the Document Image Skew Estimation Contest (DISEC) contest International Conference on Document Analysis and Recognition (ICDAR)

Specific category region proposal network for text detection in natural scene

Natural scene text usually carries considerable abstract semantic information, which is closely related to the surrounding environment. Thus, natural scene text detection plays a vital role in image content retrieval and understanding. In this study, the authors propose a novel specific category region proposal network (SCRPN) based on maximally stable extremal regions (MSER) and fully convolutional network (FCN) for natural scene text detection. First, FCN for pixel-level recognition is utilised to obtain the text saliency map and MSER is used to obtain oversegmented regions. Then, the multiple features of oversegmented regions and text saliency map are used for region aggregation. Next, single-linkage clustering method is adopted to cluster the segmentation regions to obtain a hierarchical structure of text region proposals. Finally, for the top-ranking region proposals, SCRPN built an end-to-end pipeline for scene text detection directly. Experiments on street view text and international conference on document analysis and recognition (ICDAR) 2013 have demonstrated the effectiveness of SCRPN for generating the text proposals. SCRPN could work with various two-stage text detection networks; thus, faster region convolutional neural network was used as the text detection framework to evaluate the performance of SCRPN in the ICDAR 2015 and MSRA-TD500 benchmarks. The experimental results confirmed that SCRPN makes text detection more robust in complex scenarios.

KuroNet: Regularized Residual U-Nets for End-to-End Kuzushiji Character Recognition

Kuzushiji, a cursive writing style, had been used in Japan for over a thousand years starting from the eighth century. Over 3 million books on a diverse array of topics, such as literature, science, mathematics and even cooking are preserved. However, following a change to the Japanese writing system in 1900, Kuzushiji has not been included in regular school curricula. Therefore, most Japanese natives nowadays cannot read books written or printed just 150 years ago. Museums and libraries have invested a great deal of effort into creating digital copies of these historical documents as a safeguard against fires, earthquakes and tsunamis. The result has been datasets with hundreds of millions of photographs of historical documents which can only be read by a small number of specially trained experts. Thus there has been a great deal of interest in using machine learning to automatically recognize these historical texts and transcribe them into modern Japanese characters. Our proposed model KuroNet (which builds on Clanuwat et al. in International conference on document analysis and recognition (ICDAR), 2019) outperforms other model for Kuzushiji recognition. In this paper, KuroNet achieves higher accuracy while still recognizing entire pages of text using the residual U-Net architecture from adding more regularization. We also explore areas where our system is limited and suggests directions for future work.

Stroke Sequence-Dependent Deep Convolutional Neural Network for Online Handwritten Chinese Character Recognition.

We propose a novel model, called stroke sequence-dependent deep convolutional neural network (SSDCNN), which uses the stroke sequence information and eight-directional features of Chinese characters for online handwritten Chinese character recognition (OLHCCR). SSDCNN learns the representation of OLHCCs by incorporating the natural sequence information of the strokes. Furthermore, it naturally incorporates the eight-directional features. First, SSDCNN inputs the stroke sequence and transforms it into stacks of feature maps following the writing order of the strokes. Second, the fixed-length, stroke sequence-dependent representations of OLHCC are derived through convolutional, residual, and max-pooling operations. Third, the stroke sequence-dependent representation is combined with the eight-directional features via a number of fully connected neural network layers. Finally, the Chinese characters are recognized using a softmax classifier. The SSDCNN is trained in two stages: 1) the whole architecture is pretrained using the training data until the performance converges to an acceptable degree. 2) The stroke sequence-dependent representation is combined with the eight-directional features by a fully connected neural network and a softmax layer for further training. The model was experimentally evaluated on the OLHCCR competition tasks of International Conference on Document Analysis and Recognition (ICDAR) 2013. The recognition error was a maximum 58.28% lower in SSDCNN than in a model using the eight-directional features alone (5.13% versus 2.14%). Owing to its high accuracy (97.86%), the proposed SSDCNN reduced the recognition error by approximately 18.0% as compared with that of the winning system in the ICDAR 2013 competition. SSDCNN integrated with an adaptation mechanism, called the SSDCNN+Adapt model, and reached a new state-of-the-art (SOTA) standard with an accuracy of 97.94%. The SSDCNN exploits the stroke sequence information to learn high-quality OLHCC representations. Moreover, the learned representation and the classical eight-directional features complement each other within the SSDCNN architecture.

Deep Neural Networks Combined with STN for Multi-Oriented Text Detection and Recognition

Developing systems for interpreting visuals, such as images, videos is really challenging but important task to be developed and applied on benchmark datasets. This study solves the very challenge by using STN-OCR model consisting of deep neural networks (DNN) and Spatial Transformer Networks (STNs). The network architecture of this study consists of two stages: localization network and recognition network. In the localization network it finds and localizes text regions and generates sampling grid. Whereas, in the recognition network, text regions will be input and then this network learns to recognize text including low resolution, curved and multi-oriented text. Deep learning-based approaches require a lot of data for training effectively, therefore, this study has used two benchmark datasets, Street View House Numbers (SVHN) and International Conference on Document Analysis and Recognition (ICDAR) 2015 to evaluate the system. The STN-OCR model achieves better results than literature on these datasets.

Dataset of Urduud1k from Natural Scenes

In latest years research has drawn attention on text analysis in natural scenes. Databases play a significant part in the efficiency assessment of the algorithm for text recognition. A data set of natural scene text images in six distinct languages have recently been released in an International Conference on Document Analysis and Recognition (ICDAR).This dataset is for multi-languages except Urdu. In the natural images of the Urdu scene, there is an absence of a conventional Urdu text database. This research therefore mainly aims to build a database for Urdu text in natural scenes. The dataset is very large because there are 10 distinct cameras with distinct resolution, distinct angles and distinct range requirements for each picture captured by distinct light zone. The dataset comprises of Urdu words, ligatures and characters in natural scenes. The dataset contains 16k images of words, 32k ligatures and characters images. This dataset contains 1kimagesincluding signboard, a name of the store, banners and so on. In addition, the Urdu dataset is contrasted with the current data set including ICRAR 2003, ARASTI, Chars 74k, etc. The dataset includes many images from the natural scene so it can be used in natural environments to identify Urdu text.

An Enhanced MSER Pruning Algorithm for Detection and Localization of Bangla Texts from Scene Images

Text detection and localization have great importance for content based image analysis and text based image indexing. The efficiency of text recognition depends on the efficiency of text localization. So, the main goal of the proposed method is to detect and localize text regions with high accuracy. To achieve this goal, a new and efficient method has been introduced for localization of Bangla text from scene images. In order to improve precision and recall as well as f-measure, Maximally Stable Extremal Region (MSER) based method along with double filtering techniques have been used. As MSER algorithm generates many false positives, we have introduced double filtering method for removing these false positives to increase the f-measure to a great extent. Our proposed method works at three basic levels. Firstly, MSER regions are generated from the input color image by converting it into gray scale image. Secondly, some heuristic features are used to filter out most of the false positives or non-text regions. Lastly, Stroke Width Transform (SWT) based filtering method is used to filter out remaining non-text regions. Remaining components are then grouped into candidate text regions marked by bounding box over each region. As there is no benchmark database for Bangla text, the proposed method is implemented on our own prepared database consisting of 200 scene images of Bangla texts and has got prominent performance. To evaluate the performance of our proposed approach, we have also tested the proposed method on International Conference on Document Analysis and Recognition( ICDAR) 2013 benchmark database and have got a better result than the related existing methods.

Text segmentation using superpixel clustering

Text segmentation is important for text image analysis and recognition; however, it is challenging due to noise and complex background in natural scenes. Superpixel-based image representation can enhance robustness to noise and local disturbances, but conventional superpixel algorithms are difficult to obtain the complete stroke regions and accurate boundaries for text images. In this study, a text segmentation method based on superpixel clustering is proposed. First, to generate accurate superpixels for text images, an adaptive simple linear iterative clustering-based text superpixel generation algorithm is proposed. The adaptive superpixel size and compactness are calculated to enhance boundary adherence. Second, to increase the complete coverage of strokes from superpixels, superpixel clustering merges homogeneous superpixels into larger regions for both strokes and the background. A modified density-based spatial clustering of applications with noise is proposed. Finally, stroke superpixel verification assigns each region to a stroke or to the background and the text segmentation result is obtained. The proposed method shows promising robustness to noise and complex background textures. Experimental results on the Korea Advanced Institute of Science and Technology (KAIST) scene text dataset, International Conference on Document Analysis and Recognition (ICDAR) 2003 natural scene text image dataset and Street View Text dataset verify that this method is effective and significantly outperforms existing methods.

Mean-Shift segmentation and PDE-based nonlinear diffusion: toward a common variational framework for foreground/background document image segmentation

The presence of noise in images of degraded documents limits the direct application of segmentation approaches and can lead to the presence of a number of different artifacts in the final segmented image. A possible solution is the integration of a pre-filtering step which may improve the segmentation quality through the reduction of such noise. This study demonstrated that combining the Mean-Shift clustering algorithm and the tensor-driven diffusion process into a joint iterative framework produced promising results. For instance, this framework generates segmented images with reduced edge and background artifacts when compared to results obtained after applying each method separately. This improvement is explained by the mutual interaction of global and local information, introduced, respectively, by the Mean-Shift and the anisotropic diffusion. Another point of note is that the anisotropic diffusion process smoothed images while preserving edge continuities. The convergence of this framework was defined automatically under a stopping criterion not previously defined when the diffusion process was applied alone. To obtain a fast convergence, the common framework utilizes the speedup algorithm of the Fukunaga and Hostetler Mean-Shift formulation already proposed by Lebourgeois et al. (International Conference on Document Analysis and Recognition (ICDAR), pp 52–56, 2013). This new variant of the Mean-Shift algorithm produced similar results to the original one, but ran faster due to the application of the integral volume. The first application of this framework was document ink bleed-through removal where noise is stemmed from the interference of the verso side on the recto side, thus perturbing the legibility of the original text. Other categories of images could also be subjected to the proposed framework application.

TEXT SIGNAGE RECOGNITION IN ANDROID MOBILE DEVICES

This study presents a Text Signage Recognition (TSR) model in Android mobile devices for Visually Impaired People (VIP). Independence navigation is always a challenge to VIP for indoor navigation in unfamiliar surroundings. Assistive Technology such as Android smart devices has great potential to assist VIPs in indoor navigation using built-in speech synthesizer. In contrast to previous TSR research which was deployed in standalone personal computer system using Otsu’s algorithm, we have developed an affordable Text Signage Recognition in Android Mobile Devices using Tesseract OCR engine. The proposed TSR model used the input images from the International Conference on Document Analysis and Recognition (ICDAR) 2003 dataset for system training and testing. The TSR model was tested by four volunteers who were blind-folded. The system performance of the TSR model was assessed using different metrics (i.e., Precision, Recall, F-Score and Recognition Formulas) to determine its accuracy. Experimental results show that the proposed TSR model has achieved recognition rate satisfactorily.