Text Images Research Articles

Scene Text Image Superresolution Through Multiscale Interaction of Structural and Semantic Priors

Scene Text Image Superresolution Through Multiscale Interaction of Structural and Semantic Priors

PERFORMANCE AS A COMMUNICATIVE STRATEGY OF MODERN DRAMA

Emphasized demonstrativeness, which is embodied and manifested in various spheres of human life, is a performance. Accordingly, a performative text is a self-representative text actualized in the process of communication precisely through action, a way of thinking, a director’s vision, and acting. The visual component of the text is defined as a constituent component of the poetics of the work, a gender-genre marker, according to the elements of the external composition. The usual classical image of a dramatic text in modern Ukrainian playwrights takes on the features of epic genres, which represent a different type and way of thinking, combining the narrator and the character in one person. Performance as a communicative strategy of modern Ukrainian drama is ensured by the transition to the narrative non-linearity of the character-narrative organization of the dialogue, the retrospective organization of the action, and the construction of the monologue based on associations of events, intimate-personal connections and perceptions, where the monologue is the real action of the character. The demonstrativeness and creolized nature of such text gives us the right to define such texts as “action art”.

Cross-modal semantic aligning and neighbor-aware completing for robust text–image person retrieval

Most existing text–image person re-identification (TIReID) methods are performed in an ideal environment where both image and text instances are fully intact and identity annotated. However, in real-world open environments, these ideal assumptions often cannot be satisfied. In this study, we are the first to explore how to enhance the robustness of the TIReID model to better adapt to open environments, focusing on two aspects: 1) unlabeled multi-modal data, and 2) incomplete multi-modal data. Here, unlabeled multi-modal data faces two key issues: intra-class variations and cross-modal aligning problems, while incomplete data leads to data semantic inconsistency problems and severe performance degradation. In order to tackle the above issues, we propose a novel Cross-modal Semantic Aligning and Neighbor-aware Completing (CANC) method for robust text–image person re-identification (RTIReID). Specifically, to tackle the intra-class variation problem, we propose intra-view prototype contrastive matching for image and text modalities, which helps the model learn intra-modal discriminative representations while increasing inter-class distance and suppressing intra-class variation. Furthermore, to resolve the cross-modal alignment problem, we propose the cross-view instance projection matching to establish the relationship between different modal representations. Additionally, regarding how to effectively handle incomplete samples, we introduce nearest neighbor consistent completion, which can restore high-quality completion features. We adopt CLIP-ViT and CLIP-Xformer as image and text encoders respectively, and Rank-k metrics (k=1,5,10) as primary evaluation metrics. Extensive experiments on several public datasets demonstrate that our method can effectively address TIReID in open environments, consistently surpassing several SOTA TIReID methods.

Key Information Extraction and Recognition from Rich Text Images

Key information extraction and recognition from rich text images are crucial for various applications. There are two main tasks involved in this process: Line Item Recognition (LIR) and Key Information Localization and Extraction (KILE). LIR aims at identifying and interpreting data line items in a document. The essential information in each line item is then classified or extracted, a task known as KILE. A widely used approach for this problem is sequence based, which relies on the generalization of a language model and requires a significant amount of training time. We present an effective and reliable solution to the problem by using RoBERTa, a transformer model trained on a large corpus, along with the LION optimizer to improve the training process. A comprehensive evaluation was conducted on two different benchmarks, emphasizing two different languages, English and Vietnamese. Experimental results on DocILE indicate that the proposed framework significantly improves the KILE task with a 7.24% increase in accuracy compared to the baseline and also enhances the correct recognition rate at the LIR stage. On MCOCR, the method achieved a Character Error Rate (CER) of 28.6%, which is competitive with the state-of-the-art on this dataset.

A Pix2Pix Architecture for Complete Offline Handwritten Text Normalization.

In the realm of offline handwritten text recognition, numerous normalization algorithms have been developed over the years to serve as preprocessing steps prior to applying automatic recognition models to handwritten text scanned images. These algorithms have demonstrated effectiveness in enhancing the overall performance of recognition architectures. However, many of these methods rely heavily on heuristic strategies that are not seamlessly integrated with the recognition architecture itself. This paper introduces the use of a Pix2Pix trainable model, a specific type of conditional generative adversarial network, as the method to normalize handwritten text images. Also, this algorithm can be seamlessly integrated as the initial stage of any deep learning architecture designed for handwritten recognition tasks. All of this facilitates training the normalization and recognition components as a unified whole, while still maintaining some interpretability of each module. Our proposed normalization approach learns from a blend of heuristic transformations applied to text images, aiming to mitigate the impact of intra-personal handwriting variability among different writers. As a result, it achieves slope and slant normalizations, alongside other conventional preprocessing objectives, such as normalizing the size of text ascenders and descenders. We will demonstrate that the proposed architecture replicates, and in certain cases surpasses, the results of a widely used heuristic algorithm across two metrics and when integrated as the first step of a deep recognition architecture.

Deep RegNet-150 architecture for single image super resolution of real-time unpaired image data

Single Image Super-Resolution (SISR) is a fundamental computer vision task aimed at enhancing the spatial resolution and quality of low-resolution images. In recent years, deep learning techniques have revolutionized the field of SISR, enabling remarkable advancements in image enhancement. Typically, the Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs) based deep learning techniques are employed for the SISR. However, the existing CNN and GAN architectures lag in providing maximum performance in terms of SSIM and PSNR. In this paper, a regression deep learning architecture is designed for the SISR. The proposed Deep Regression Network with 150 (Deep RegNet-150) is designed as a U-net. The proposed Deep RegNet-150 includes up-sampling and down-sampling portions with Residual Channel Attention Block (RCAB) for SISR. The RCAB acts as the building block of image super resolution (SR) by including ReLU and Conv layer. The performance of the proposed technique is evaluated on different image datasets, including a text image dataset and an unpaired dataset. The performance in terms of SSIM and PSNR for 4X and 8X SR is evaluated. The proposed Deep RegNet-150 attained maximum PSNR / SSIM values of 43.99 / 0.989 and 42.73 / 0.987 for 4X and 8X SR of the benchmark dataset, respectively. The 4X and 8X SR of unpaired datasets attained 46.13 / 0.991 and 45.97 / 0.990 of PSNR /SSIM, respectively. Ultimately, it proves that the Deep RegNet-150 is an effective deep learning architecture for the SISR.

SterAF: A Scene Text Recognizer with Appearance-Flow Rectification

As the demand for recognizing irregular text in natural scenes increases, people are increasingly realizing the value of such applications, such as license plate recognition systems, image search, handwriting recognition, and autonomous driving, which are profoundly changing our lives in the field of text recognition. Recent studies have shown that the recognition of curved text and perspective text has become an important challenge in the field of text recognition, and the correction of curved text is a key step to achieve accurate recognition. However, current methods use strained text image correction methods, resulting in poor recognition accuracy when recognizing curved text. Therefore, we propose an end-to-end framework called Scene Text Recognizer with Appearance-Flow rectification (SterAF), which includes a correction network and a recognition network. Specifically, the framework’s steps are as follows: first, the input text image is deformed through an appearance flow-based correction network to adaptively warp the text image, to prevent irregular and unnatural deformations of the text image. Second, a sequence-to-sequence recognition network predicts the sequence of characters in the corrected text image to accurately recognize the text in the image. Through subjective and objective experiments, our SterAF model has shown excellent performance in both qualitative and quantitative experiments.

Poetic Images in (The House of Female Reciters) By the Poet Sadiq Al-Tarihi

It has become clear to us that the simile image is the dominant image in the poet’s collection of poetry (The House of Female Reciters) and that the metonymic image is ranked second in terms of the intensity of its presence after the simile image. The Qur’anic textual reference represents the main tributary to the formation of the image in the House of Female Reciters, while the literary textual image comes in first place. Second after the Qur’an in terms of its presence. Traditional and contemporary poetic attempts represented a clear presence in (The House of Female Reciters), as the textual reference enriched its texts.

A Study on Handwritten Text Recognition Classification using Diverse Deep Learning Techniques and Computation of CTC Loss

Handwriting recognition encompasses the conversion of hand-written text images into digital text, where input images yield predicted textual output. Optical Character Recognition (OCR) technology has conventionally fulfilled this role. With surging mobile phone usage, leveraging text detection via mobile cameras gains significance in fields like medical script processing and exam evaluation. To enhance image quality, noise reduction techniques like binarization and thresholding are applied. Image processing entails letter segmentation and extraction. In this paper, we propose a neural network classifier model amalgamating Convolution Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. Leveraging an existing feature dataset, the model is trained. Input images are evaluated through the neural network, yielding recognized words in a text document format. Ultimately, the predicted output is derived via Connectionist Temporal Classification (CTC)- based loss computation. Keywords: Handwritten Recognition, Deep Learning Techniques, Optical Character Recognition.

Blind deblurring text images via Beltrami regularization

This article proposes a blind image deblurring model based on Beltrami regularization. The existence and uniqueness of the Beltrami model are proved, and we perform a theoretical analysis of the convergence and error estimation of the algorithm for solving the Beltrami regularization model. Meanwhile, we apply a half-quadratic splitting algorithm to solve it and obtain the blur kernel of the blurred image. Furthermore, we get a clean image through some non-blind deblurring algorithms. Finally, we compare this with the state-of-art methods, and our algorithm has specific performance improvements and advantages in the text images.© 2024 Elsevier Ltd. All rights reserved.

Scene text image super-resolution using multi-scale convolutional neural network with skip connections

Scene text image super-resolution using multi-scale convolutional neural network with skip connections

Toyok (Tuyugou) Cave 20: A Pure Land Cave Temple in the Desert with the Earliest Illustrations of the Visualization Sūtra

In this paper, I examine the iconography, religious function and text–image relationship of one of the earliest illustrations of the apocryphal Visualization Sūtra, namely, the mural painting on the left (south) wall of Cave 20 at the Toyok Grottoes in Turfan created in the late sixth century. I study this mural with a structuralist approach to situate it within the overall pictorial program of the cave temple. I argue that this wall painting was designed as a visual and verbal cue to assist the meditating monks in situ to separately visualize the different individual visions constituting the “Thirteen Visualizations” taught in the sutra, excluding the five visions directly related to the Amitāyus Triad, namely, the eighth, ninth, tenth, eleventh and thirteenth visualizations. These five visualizations were likely facilitated by icons of the Amitāyus Triad placed in the center of the cave. The meditators started from the contemplations of the impure represented on the right (north) wall as preparational practices to eliminate their sins, improve their karma and thereby enhance their spiritual purity to a level appropriate for performing the Pure Land visualizations. They then turned to the left wall to perform the Pure Land visualizations represented there that end with the twelfth visualization: imagining oneself being reborn in Sukhāvatī. Continuing from the twelfth visualization, the meditators, in a state of meditational concentration, turned their faces to the rear wall to enter the Western Pure Land via the various lotus ponds represented on the rear wall.

Islamic marketing of conventional banks: bridging managers’ and clients’ perceived gaps

Purpose While the Noble Quran dictates the prohibition of interest, conventional banks promote Islamic banking by opening Islamic banking windows. Against this backdrop, this study aims to investigate the perceived gaps between managers and clients in Islamic marketing and banking, focusing on conventional banks’ Islamic banking windows. Design/methodology/approach Guided by a qualitative approach, semi-structured personal interviews and observations served as the data collection methods, involving 25 banks and 50 respondents in 3 different districts, namely, Shirajganj, Rajshahi and Chapainawabganj of Bangladesh from January to October 2023. The data were analysed using ATLAS.ti 2023 to explore codes and quotations derived from 14 interview questions. Further, ATLAS.ti 2023 facilitated synthesizing content, concepts, code occurrence, network analysis and thematic analysis. Findings Islamic and non-Islamic banks use Quranic verses, hadiths (prophetic traditions), images of mosques, the Kaaba and Arabic texts as Islamic marketing tools. These spiritual, divine and prescriptive tools are associated with Islamic banking. However, conventional banks receive criticism for having separate Islamic banking windows to serve religiously conscious clients, which generates tension among clients and bank managers. Practical implications The findings can theoretically assist academics in examining conventional banks’ Islamic marketing and banking practices, opening Islamic banking windows. Importantly, Shariah boards can play policy roles in safeguarding the function of Islamic marketing and banking. Managers can use the findings to anticipate client perceptions and enhance Islamic marketing and banking strategies. Likewise, the social implications include the explicit stance of Shariah to mitigate the mixture of halal and haram banking. Originality/value This pioneering study explores the perspectives of Islamic banking windows by non-Islamic banks. The combination of Islamic marketing and banking is a noteworthy novelty in this study and deserves recognition for its unique contribution to halal marketing and finance.

MISTRA: Misogyny Detection through Text–Image Fusion and Representation Analysis

Detecting misogynous memes poses a significant challenge due to the presence of multiple modalities (image + text). The inherent complexity arises from the lack of direct correspondence between the textual and visual elements, where an image and overlaid text often convey disparate meanings. Additionally, memes conveying messages of hatred or taunting, particularly targeted towards women, present additional comprehension difficulties. This article introduces the MISTRA framework, which leverages variational autoencoders for dimensionality reduction of the large-sized image features before fusing multimodal features. The framework also harnesses the capabilities of large language models through transfer learning to develop fusion embeddings by extracting and concatenating features from different modalities (image, text, and image-generated caption text) for the misogynous classification task. The components of the framework include state-of-the-art models such as the Vision Transformer model (ViT), textual model (DistilBERT), CLIP (Contrastive Language–Image Pre-training), and BLIP (Bootstrapping Language–Image Pre-training for Unified Vision-Language Understanding and Generation) models. Our experiments are conducted on the SemEval-2022 Task 5 MAMI dataset. To establish a baseline model, we perform separate experiments using the Naive Bayes machine learning classifier on meme texts and ViT on meme images. We evaluate the performance on six different bootstrap samples and report evaluation metrics such as precision, recall, and Macro-F1 score for each bootstrap sample. Additionally, we compute the confidence interval on our evaluation scores and conduct paired t-tests to understand whether our best-performing model has significant differences from the other experiments or not. The experimental results demonstrate that the dimensionality reduction approach on multimodal features with a multilayer perceptron classifier achieved the highest performance with a Macro–F1 score of 71.5 percent, outperforming the baseline approaches in individual modalities.

Reading instruction causes changes in category-selective visual cortex

Education sculpts specialized neural circuits for skills like reading that are critical to success in modern society but were not anticipated by the selective pressures of evolution. Does the emergence of brain regions that selectively process novel visual stimuli like words occur at the expense of cortical representations of other stimuli like faces and objects? “Neuronal Recycling” predicts that learning to read should enhance the response to words in ventral occipitotemporal cortex (VOTC) and decrease the response to other visual categories such as faces and objects. To test this hypothesis, and more broadly to understand the changes that are induced by the early stages of literacy instruction, we conducted a randomized controlled trial with pre-school children (five years of age). Children were randomly assigned to intervention programs focused on either reading skills or oral language skills and magnetoencephalography (MEG) data collected before and after the intervention was used to measure visual responses to images of text, faces, and objects. We found that being taught reading versus oral language skills induced different patterns of change in category-selective regions of visual cortex, but that there was not a clear tradeoff between the response to words versus other categories. Within a predefined region of VOTC corresponding to the visual word form area (VWFA) we found that the relative amplitude of responses to text, faces, and objects changed, but increases in the response to words were not linked to decreases in the response to faces or objects. How these changes play out over a longer timescale is still unknown but, based on these data, we can surmise that high-level visual cortex undergoes rapid changes as children enter school and begin establishing new skills like literacy.

A Review on Currency Classification and Image to Text Conversion Methodologies

Currency classification and Image to Text OCR are essential technologies that find applications in various domains, including finance, retail, and automation. The approach outlined in this paper has the potential to detect currencies from multiple countries. However, for practical implementation purposes, the focus is solely on Indian paper currencies. This system offers the advantage of convenient currency checking at any time and location, leveraging Convolutional Neural Networks (CNN) for effective implementation. Extensive testing was conducted on each denomination of Indian currency, resulting in an impressive 95% accuracy rate. To further refine accuracy, a classification model was developed, incorporating all pertinent factors discussed in the paper. Notably, the unique features of paper currency play a pivotal role in the recognition process. By emphasizing these elements and harnessing CNN technology, the proposed system demonstrates significant promise in accurately detecting and validating Indian paper currencies. It stands poised to serve various applications effectively. On the other hand, Image to Text OCR focuses on extracting text from images, enabling the conversion of non- editable documents into searchable and editable formats. Both technologies contribute to automation and efficiency in handling diverse visual information. Optical Character Recognition (OCR) is a technologydesigned to recognize and interpret both printed and handwritten characters by scanning text images. This process involves segmenting the text image into regions, isolating individual lines, and identifying each character along with its spacing. After isolating individual characters from the text image, the system conducts an analysis of their texture and topological attributes. This involves examining corner points, unique characteristics of various regions within the characters, and calculating the ratio of character area to convex area Prior to initiating recognition, the system creates templates that store the distinctive features of uppercase and lowercase letters, digits, and symbols. These templates serve as reference models for comparison during the recognition phase. During recognition, the system matches the extracted character's texture and topological Features with those stored in the templates to determine the exact character. This matching process involves comparing features of the extracted character with templates of all characters, measuring similarity, and ultimately recognizing the character accurately.

Images Connect Us Together: Navigating a COVID-19 Local Outbreak in China Through Social Media Images

Social media images, curated or casual, have become a crucial component of communicating situational information and emotions during health crises. Despite its prevalence and significance in informational dissemination and emotional connection, there lacks a comprehensive understanding of visual crisis communication in the aftermath of a pandemic which is characterized by uncertain local situations and emotional fatigue. To fill this gap, this work collected 345,423 crisis-related posts and 65,376 original images during the Xi'an COVID-19 local outbreak in China, and adopted a mixed-methods approach to understanding themes, goals, and strategies of crisis imagery. Image clustering captured the diversity of visual themes during the outbreak, such as text images embedding authoritative guidelines and "visual diaries" recording and sharing the quarantine life. Through text classification of the post that visuals were situated in, we found that different visual themes highly correlated with the informational and emotional goals of the post text, such as adopting text images to convey the latest policies and sharing food images to express anxiety. We further unpacked nuanced strategies of crisis image use through inductive coding, such as signifying authority and triggering empathy. We discuss the opportunities and challenges of crisis imagery and provide design implications to facilitate effective visual crisis communication.

A real-time arbitrary-shape text detector.

It is challenging to detect arbitrary-shape text accurately and effectively in natural scenes. While many methods have been implemented for arbitrary-shape text detection, most cannot achieve real-time detection or meet practical needs. In this work, we propose a YOLOv6-based detector that can effectively implement arbitrary-shape text detection and achieve real-time detection. We include two additional branches in the neck part of the YOLOv6 network to adapt the network to text detection, and the output side uses the pixel aggregation (PA) algorithm to decouple the PA output to use it as the detection head of the model. Experiments on benchmark Total-Text, CTW1500, ICDAR2015, and MSRA-TD500 showed that the proposed method outperformed competing methods in terms of detection accuracy and running time. Specifically, our method achieved an F-measure of 84.1% at 291.8 FPS for 640 × 640 Total-Text images and an F-measure of 81.5% at 199.6 FPS for 896 × 896 ICDAR2015 incidental text images.

Image-Deblurring and Restoration: Exploring Advanced Techniques for Enhanced Visual Clarity

This study introduces an end-to-end scale-recurrent deep network for deblurring multi-modal medical images affected by motion artifacts. The network features a novel residual dense block with spatial-asymmetric attention, enhancing the recovery of salient information during medical image deblurring. Comprehensive evaluations demonstrate the method's superiority over existing techniques, showcasing its ability to remove blur without introducing visual artifacts. Integration into medical image analysis tasks, such as segmentation and detection, highlights accelerated performance. Additionally, a simple L0-regularized prior is proposed for text image deblurring, leveraging intensity and gradient properties. The method outperforms state-of-the-art text image deblurring algorithms, demonstrating effectiveness in deblurring low- illumination images without complex edge selection strategies. Our proposed end-to-end scale-recurrent deep network for medical image deblurring exhibits robust performance, outshining existing methods in both qualitative and quantitative evaluations. The inclusion of a spatial-asymmetric attention mechanism and a residual dense block enhances the network's ability to recover crucial information from blurred medical images. Furthermore, the seamless integration of our deblurring method into various medical image analysis tasks showcases its versatility and effectiveness in accelerating performance. On a different front, our contribution extends to text image deblurring, where a straightforward yet powerful L0-regularized prior based on intensity and gradient proves effective. This method not only surpasses state-of-the-art text image deblurring algorithms but also demonstrates applicability to challenging scenarios such as deblurring low-illumination images. The combined results highlight the potential impact of our approaches in advancing the quality and applicability of deblurring techniques in both medical imaging and text-related applications. Keywords—Multi-Scale, Unet, Deblurring, Fast Fourier Transform, ResBlock, Non-linear Activation Free Block, NFResnet, Charbonnier, Edge, Frequency Reconstruction.

Fashioning Landscapes with Bruno Latour’s Immutable Mobiles and Epistemological Matrix or: How We Pass on Knowledge

Landscape poetry and art is informed by place and place is understood at least in part by quantitative data about the place. This essay delineates Bruno Latour’s method of understanding how data is gathered in the field and translated into visual and textual images of place (which he calls immutable mobiles). Textual representations reduce the material specificity of place, but in return representations facilitate understanding of a place by creating greater compatibility, standardization, calculation, and circulation of knowledge. Latour explains how instruments create representations and how the instruments and texts are part of a broader implicate epistemological matrix. This matrix determines in advance how we think about, measure, see, and represent the world around us. The essay then develops how landscape poetry and art participate in the broader cultural epistemological matrix.