Stop word location and identification for adaptive text recognition

Methods for text segmentation from scene images

We present an autonomous visual landmark recognition and pose estimation algorithm designed for use in navigation of spacecraft around small asteroids. Landmarks are selected as generic points on the asteroid surface that produce strong Harris corners in an image under a wide range in viewing and illumination conditions; no particular type of morphological feature is required. The set of landmarks is triangulated to obtain a tightly fitting mesh representing an optimal low resolution model of the natural asteroid shape, which is used onboard to determine the visibility of each landmark and enables the algorithm to work with highly concave bodies. The shape model is also used to estimate the centre of brightness of the asteroid and eliminate large translation errors prior to the main landmark recognition stage. The algorithm works by refining an initial estimate of the spacecraft position and orientation. Tests with real and synthetic images show good performance under realistic noise conditions. Using simulated images, the median landmark recognition error is 2m, and the error on the spacecraft position in the asteroid body frame is reduced from 45m to 21m at a range of 2km from the surface. With real images the translation error at 8km to the surface increases from 107m to 119m, due mainly to the larger range and lack of sensitivity to translations along the camera boresight. The median number of landmarks detected in the simulated and real images is 59 and 44 respectively. This algorithm was partly developed and tested during industrial studies for the European Space Agency’s Marco Polo-R asteroid sample return mission.

In this letter, a structure filling and matching method is proposed for the 3-D reconstruction of buildings, particularly those with combinational structures, from a single high-resolution synthetic aperture radar (SAR) image. This method consists of two stages. First, the structure model of the investigated building is constructed by filling the building structure elements, i.e., walls and roofs, into an initialized empty scene. Subsequently, a series of simulated SAR images is generated by this structure model and matched with the real SAR image of the building. By finding the maximal mutual information between the simulated and real images, the optimal geometric parameters of the model are obtained, which make the output more accurate. Compared with other methods, this method can reconstruct different kinds of practical buildings and retrieve their exact geometric parameters, using only a single SAR image. The performance of this method is evaluated on real TerraSAR-X images to show its effectiveness.

Historical newspapers are invaluable resources that capture the cultural, social, economic, and political dimensions of their time, providing unique insights into past events and daily life. However, the potential of these archives remains underutilised due to the limitations of traditional research methods, which rely on laborious manual examination of individual pages or articles. Digital archives have begun to address this issue by enhancing accessibility, but the lack of effective tools for parsing and analysing the contents still restricts their utility for large-scale historical research. Optical Character Recognition (OCR) technology plays a foundational role in converting scanned images into searchable text. However, OCR alone cannot adequately handle the complexities of historical newspapers, which often feature degraded image quality, inconsistent typography, and intricate layouts. In parallel, the effective segmentation and ordering of newspaper contents is essential for generating well-structured and logically ordered data from specific newspaper sections. This paper focuses on the development of a pipeline that addresses these challenges, using the daily newspaper Le Sémaphore de Marseille as a case study (35,703 issues covering the period 1827 to 1944). The proposed system integrates layout analysis, OCR, and Generative AI-driven information extraction techniques to extract specific data elements from scanned images, namely ship arrivals data, and to store them in a machine-readable format for further analysis. This well-structured output enables historians to analyse historical trade patterns, economic trends, and regional interactions more efficiently and at a much larger scale. An evaluation of the key pipeline components demonstrates the effectiveness of the system, achieving an F1-score of 96% in paragraph segmentation and information extraction, while also highlighting the need for further improvement in the critical OCR component.

Optical Character Recognition (OCR) is a technique that generates text from an image. Recognizing the importance of OCR in real-world settings, a plethora of techniques have been developed for Western, as well as Asian languages. Urdu is a prominent South Asian language and a number of different solutions for Urdu OCR have been proposed. However, fewer attempts have been made to develop end-to-end deep learning-based solutions for recognizing printed Urdu text. Furthermore, several benchmark corpora for Urdu OCR have been developed that can be used for training and evaluation of different OCR techniques. However, there are a number of limitations of the existing Urdu corpora: firstly, most of them have either character or word or text images, which are usually rendered using only a single font, Nastaleeq. Secondly, the volume of the existing datasets is so small that it is not suitable for working with the deep-learning techniques that have achieved groundbreaking results for OCRs. To that end, in this study, we have proposed a very large Multi-level and Multi-script Urdu corpus (MMU-OCR-21). It is the largest-ever Urdu corpus of printed text that is effectively suitable to work with deep learning techniques. In total, the corpus is composed of over 602,472 images, including text-line and word images in three prominent fonts, and their respective ground truth. Also, we have performed experiments using multiple state-of-the-art deep learning techniques for text-line and word level images.

An efficient word spotting framework is proposed to search text in scanned books. The proposed method allows one to search for words when optical character recognition (OCR) fails due to noise or for languages where there is no OCR. Given a query word image, the aim is to retrieve matching words in the book sorted by the similarity. In the offline stage, SIFT descriptors are extracted over the corner points of each word image. Those features are quantized into visual terms (visterms) using hierarchical K-Means algorithm and indexed using an inverted file. In the query resolution stage, the candidate matches are efficiently identified using the inverted index. These word images are then forwarded to the next stage where the configuration of visterms on the image plane are tested. Configuration matching is efficiently performed by projecting the visterms on the horizontal axis and searching for the Longest Common Subsequence (LCS) between the sequences of visterms. The proposed framework is tested on one English and two Telugu books. It is shown that the proposed method resolves a typical user query under 10 milliseconds providing very high retrieval accuracy (Mean Average Precision 0.93). The search accuracy for the English book is comparable to searching text in the high accuracy output of a commercial OCR engine.

In recent years, automatic target recognition (ATR) based on deep learning has achieved great success in RGB field, which has huge data support. However, due to the confidentiality of military targets, weather constraints, and high shooting costs, it is difficult to obtain a large number of real IR images which leads to the performance degradation of deep learning algorithms in IR field. This paper discusses the method of using simulation IR images as training set to get rid of dependence on the real image. However, there are still great differences between the original simulated image and the real image, which leads to many defects when using the original simulated image for training. Therefore, in this paper, we use cycleGAN to convert the original simulation image into the intermediate image closer to the real image based on generative adversarial networks (GAN). Finally, the effectiveness of this method is proved by experiments.

Hyperspectral image provides a way to analyze component and the fraction by spectral unmixing. The classical method is Least Squares algorithm. But it is defined without any constraint on the abundance. The Fully constrained Least Squares (FCLS) algorithm allows resulted abundances satisfying sum-to-one and non-negative, which is practical. It has been widely used till now because it may work quickly with a small unmxing error. But when the number of endmembers is large, it gets much slow. And unmxing error is high if the endmembers are not found out completely. As for improving the speed of the algorithm, there have been many research works. But fewer algorithms are developed for increasing accurateness. Here in this paper, a primal-dual interior-point method is tested for the fully constraint unmixing with the linear unmixing model being modified slightly. A dual problem is established firstly, and then the process of the algorithm is designed to decrease distance between original problem and the dual problem based on result of FCLS. In order to promise high unmxing accurateness, the unmxing error of resulted abundances is added into the optimizing goal. Several algorithms of fully constraint abundance estimation are tested on both simulation and real hyperspectral images. The experimental results show that the prompted algorithm may find the optimal abundance for existing endmembers even though the endmember set is not complete. Accurateness of unmxing is increased, especially on the real image.

A robust star identification algorithm with star shortlisting

A hierarchical approach to recognition of handwritten Bangla characters

Text recognition in natural scene images is a challenging problem in computer vision. Different than the optical character recognition (OCR), text recognition in natural scene images is more complex due to variations in text size, colors, fonts, orientations, complex backgrounds, occlusion, illuminations and uneven lighting conditions. In this paper, we propose a segmentation-free method based on a deep convolutional recurrent neural network to solve the problem of cursive text recognition, particularly focusing on Urdu text in natural scenes. Compared to the non-cursive scripts, Urdu text recognition is more complex due to variations in the writing styles, several shapes of the same character, connected text, ligature overlapping, stretched, diagonal and condensed text. The proposed model gets a whole word image as an input without pre-segmenting into individual characters, and then transforms into the sequence of the relevant features. Our model is based on three components: a deep convolutional neural network (CNN) with shortcut connections to extract and encode the features, a recurrent neural network (RNN) to decode the convolutional features, and a connectionist temporal classification (CTC) to map the predicted sequences into the target labels. To increase the text recognition accuracy further, we explore deeper CNN architectures like VGG-16, VGG-19, ResNet-18 and ResNet-34 to extract more appropriate Urdu text features, and compare the recognition results. To conduct the experiments, a new large-scale benchmark dataset of cropped Urdu word images in natural scenes is developed. The experimental results show that the proposed deep CRNN network with shortcut connections outperform than other network architectures. The dataset is publicly available and can be downloaded from <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://data.mendeley.com/datasets/k5fz57zd9z/1</uri> .

Handwritten word image matching based on Heat Kernel Signature

Although font recognition is a fundamental issue in the field of document analysis and recognition, it was usually ignored in the past. With the development of optical character recognition (OCR), font recognition becomes more and more important. This paper proposed a well-designed convolutional neural network (CNN) architecture for traditional Mongolian font recognition by means of a single word. To be specific, the whole word image is regarded as input of CNN. Hence, the word images should be normalized into the same size before being inputted into CNN. By comparison, an appropriate aspect ratio for the traditional Mongolian word images has been determined. Experimental results demonstrate that the proposed CNN architecture outperforms three classic CNN architectures, including LeNet-5, AlexNet and GoogLeNet. Therefore, the proposed CNN is much more suitable for the task of the traditional Mongolian font recognition in the way of a single word.

Authors' names are a critical bibliographic element when searching or browsing academic articles stored in digital libraries. However, extracting such bibliographic data from printed documents requires human intervention; it is therefore not cost-effective, even using various document image-processing techniques such as Optical Character Recognition (OCR). In this paper, we describe an automatic authors' names extraction method for academic articles scanned with OCR mark-up. The proposed method first extracts authors' blocks, which include assumed author/delimiter characters based on layout analysis, and then uses a specifically designed Hidden Markov Model (HMM) for labeling the unsegmented character strings in the block as those of either an author or a delimiter. We applied the proposed method to Japanese academic articles. Results of these experiments showed that the proposed method correctly extracted more than 99% of authors' blocks with manual tuning; the proposed HMM correctly labeled more than 95% of the author name strings.

Computationally intelligent recognition of characters and symbols addresses a wide range of applications including foreign language translation and chemical formula identification. The combination of intelligent learning and optimization algorithms with layered neural structures offers powerful techniques for character recognition. These techniques were originally developed by Sandia National Laboratories for pattern and spectral analysis; however, their ability to optimize vast amounts of data make them ideal for character recognition. An adaptation of the Neural Network Designer soflsvare allows the user to create a neural network (NN_) trained by a genetic algorithm (GA) that correctly identifies multiple distinct characters. The initial successfid recognition of standard capital letters can be expanded to include chemical and mathematical symbols and alphabets of foreign languages, especially Arabic and Chinese. The FIN model constructed for this project uses a three layer feed-forward architecture. To facilitate the input of characters and symbols, a graphic user interface (GUI) has been developed to convert the traditional representation of each character or symbol to a bitmap. The 8 x 8 bitmap representations used for these tests are mapped onto the input nodes of the feed-forward neural network (FFNN) in a one-to-one correspondence. The input nodes feed forward into a hidden layer, and the hidden layer feeds into five output nodes correlated to possible character outcomes. During the training period the GA optimizes the weights of the NN until it can successfully recognize distinct characters. Systematic deviations from the base design test the network's range of applicability. Increasing capacity, the number of letters to be recognized, requires a nonlinear increase in the number of hidden layer neurodes. Optimal character recognition performance necessitates a minimum threshold for the number of cases when genetically training the net. And, the amount of noise significantly degrades character recognition efficiency, some of which can be overcome by adding noise during training and optimizing the form of the network's activation fimction.