Document Image Research Articles

Enhancement of Degraded Historical Document Images for Binarization

Enhancement of Degraded Historical Document Images for Binarization

FDM: Document image seen-through removal via Fuzzy Diffusion Models

While scanning or shooting a document, factors like ink density and paper transparency may cause the content from the reverse side to become visible through the paper, resulting in a digital image with a ‘seen-through’ phenomenon, which will affect practical applications. In addition, document images can be affected by random factors during the imaging process, such as differences in the performance of camera equipment and variations in the physical properties of the document itself. These random factors increase the noise of the document image and may cause the seen-through phenomena to become more complex and diverse. To tackle this issue, we propose the Fuzzy Diffusion Model (FDM), which combines fuzzy logic with diffusion models. It effectively models complex seen-through effects and handles uncertainties in document images. Specifically, we gradually degrade the original image with mean-reverting stochastic differential equation(SDE) to transform it into seen-through mean state with fixed Gaussian noise version. Following this, fuzzy operations are introduced into the noise network. Which helps the model better learn noise and data distributions by reasoning about the affiliation relationship of each pixel point through fuzzy logic. Eventually, in the reverse process, the low-quality image is gradually restored by simulating the corresponding reverse-time SDE. Extensive quantitative and qualitative experiments conducted on various datasets demonstrate that the proposed method significantly removes the seen-through effects and achieves good results under several metrics. The proposed FDM effectively solves the seen-through effects of document images and obtains better visual quality.

3차원 재구성과 텍스트 경계 보완에 기반한 문서 이미지 평판화 기법

3차원 재구성과 텍스트 경계 보완에 기반한 문서 이미지 평판화 기법

Appearance Enhancement for Camera-Captured Document Images in the Wild

Appearance Enhancement for Camera-Captured Document Images in the Wild

Mapping of Land Use Suitability for Binjai City in 2020-2024 Using SPOT-6 Imagery

Indonesia is a developing country that experiences high population growth every year. Population growth has an impact on the high demand for land use. Land use is often not in harmony with population growth, so land use does not align with regional spatial conditions. This study aims to determine land use in 2021 in Binjai City based on the 2020-2040 Regional Spatial Plan (RTRW) using Spot 6 imagery and determine the suitability of 2021 land use in Binjai City based on the 2020-2020 Regional Spatial Plan (RTRW) 2040 using Image Spot 6. This study is located in Binjai City, administratively flanked by two significant regencies: Langkat Regency and Deli Serdang Regency. The population of this research is the City of Binjai, with an area of 90.23 km2. Data collection techniques are documentation study techniques, image interpretation, and field observations. Data analysis was carried out in a qualitative descriptive manner. The results of the study show that Land use in Binjai City in 2021 includes an agricultural area of 52.93 km2 with a percentage of 58.66%, then a residential area of 33.85 km2 with a percentage of 37.63%, area conservation has an area of 1.58 km2 with a percentage of 1.75%, river bodies have an area of 0.90 km2 with a percentage of 1%, local protected areas have an area of 0.38 km2 with a percentage of 0.39%, defense and security areas have an area of 0.45 km2 with the percentage of 0.42%, industrial designation area of 0.12 km2 with a percentage of 0.15%, tourism area of 0.01 km2 with a percentage of 0.01%, and lakes/situ area of 0.01% 0.01 km2 with a percentage of 0.01%. The suitability of land use in 2021 with the Binjai City Spatial Plan (RTRW) for 2020-2040 is an area of 43.64 km2 with a percentage of 48.36% which is suitable while that which is not suitable is 46, 59 km2 with a percentage of 51.64%.

Histogram-based global thresholding method for image binarization

One of the most fundamental issues in image processing is the thresholding (binarization) method. This method is generally used for segmenting regions with different homogeneity in grayscale images. In other words, it performs clustering based on the intensity levels of pixels in an image histogram. This paper presents a new and effective approach to the global thresholding method of grayscale images. In the proposed method, alpha and beta regions are determined using the mean and standard deviation values of an image histogram. The optimum threshold value is obtained by calculating the average of gray-scale values of the alpha and beta regions. The experiments were carried out on three different image sets to demonstrate the effectiveness of the thresholding method. The result of experimental studies show that the proposed method achieves promising performance compared to many traditional, state-of-the-art thresholding and document binarization methods performed in H-DIBCO'14 (Document Image Binarization Competition), Human HT29 colon-cancer cells (BBBC008) and C. elegans live/dead assay (BBBC010) datasets based on various evaluation criteria.

A Review of Document Binarization: Main Techniques, New Challenges, and Trends

Document image binarization is a challenging task, especially when it comes to text segmentation in degraded document images. The binarization, as a pre-processing step of Optical Character Recognition (OCR), is one of the most fundamental and commonly used segmentation methods. It separates the foreground text from the background of the document image to facilitate subsequent image processing. In view of the different degradation degrees of document images, researchers have proposed a variety of solutions. In this paper, we have summarized some challenges and difficulties in the field of document image binarization. Approximately 60 methods documenting image binarization techniques are mentioned, including traditional algorithms and deep learning-based algorithms. Here, we evaluated the performance of 25 image binarization techniques on the H-DIBCO2016 dataset to provide some help for future research.

DocNLC: A Document Image Enhancement Framework with Normalized and Latent Contrastive Representation for Multiple Degradations

Document Image Enhancement (DIE) remains challenging due to the prevalence of multiple degradations in document images captured by cameras. In this paper, we respond an interesting question: can the performance of pre-trained models and downstream DIE models be improved if they are bootstrapped using different degradation types of the same semantic samples and their high-dimensional features with ambiguous inter-class distance? To this end, we propose an effective contrastive learning paradigm for DIE — a Document image enhancement framework with Normalization and Latent Contrast (DocNLC). While existing DIE methods focus on eliminating one type of degradation, DocNLC considers the relationship between different types of degradation while utilizing both direct and latent contrasts to constrain content consistency, thus achieving a unified treatment of multiple types of degradation. Specifically, we devise a latent contrastive learning module to enforce explicit decorrelation of the normalized representations of different degradation types and to minimize the redundancy between them. Comprehensive experiments show that our method outperforms state-of-the-art DIE models in both pre-training and fine-tuning stages on four publicly available independent datasets. In addition, we discuss the potential benefits of DocNLC for downstream tasks. Our code is released at https://github.com/RylonW/DocNLC

An Unsupervised Character Recognition Method for Tibetan Historical Document Images Based on Deep Learning

As there is a lack of public mark samples of Tibetan historical document image characters at present, this paper proposes an unsupervised Tibetan historical document character recognition method based on deep learning (UD-CNN). Firstly, using the Tibetan historical document character component, the Tibetan historical document character sample data set is constructed for model-aided training. Then, the character baseline information is introduced, and a fine-grained feature learning strategy is proposed. For the samples above and below the baseline, the Up-CNN recognition model and Down-CNN recognition model are established. The convolution neural network model is trained and optimized for the samples above and below the baseline, respectively, to improve the recognition accuracy. The experimental results show that the proposed method obviously affects the unmarked character classification and recognition of real Tibetan historical document images. The recognition rate of Top5 can reach 92.94%, and the recognition rate of Top1 can be increased from 82.25% to 87.27% using the CNN model only.

Proposal of ASLA Which Is a Segmentation and Labeling Tool for Document Images Based on Deep Learning

Proposal of ASLA Which Is a Segmentation and Labeling Tool for Document Images Based on Deep Learning

Assemblages as Ecologies

In this essay I draw on my art research and installation ‘Earthly Bodies, Subterranean Rhythms’ to posit assemblages as ecologies: platforms where specific agencies and interactions occur at different scales, including microscopic levels. Based on documentation images of my creative process, I delve into my experience exploring sculptural building and creative possibilities in more-than-human collaboration. This research involved observing subterranean bodies and temporalities, rhizomatic growth and mycorrhizal interactions as well as working in interspecies cooperation with oyster mushroom mycelium and wheatgrass roots. As spatial practices, architecture and sculpture have developed specific building methods. I expand on assemblage, the building method I worked with, seeking to articulate three lenses: sculptural, material and philosophical. Technically, an assemblage is a piece made by bringing together disparate elements. In this case it consisted mainly of ceramics, different earthly substrates, mycelium spawn, soil and seeds. Borrowing from Bennet’s ‘Vibrant Matter’ (2010) and Tsing’s ‘The Mushroom at the End of the World’ (2015) I reflect on living sculptures as interspecies assemblages: spaces of collaboration, cooperation and contamination between living and non-living bodies, matter and forces, human and more-than human agencies. From this perspective, assemblages are more than the sum of their parts as they have the capacity to re-make and transform us through encounters. A polyphony of worlding processes makes the collective architecture we live in: our shared world. Working in interspecies assemblages may enable more respectful and enjoyable ways of collective dwelling.

DNN-HHOA: Deep Neural Network Optimization-Based Tabular Data Extraction from Compound Document Images

Text information extraction from a tabular structure within a compound document image (CDI) is crucial to help better understand the document. The main objective of text extraction is to extract only helpful information since tabular data represents the relation between text lying in a tuple. Text from an image may be of low contrast, different style, size, alignment, orientation, and complex background. This work presents a three-step tabular text extraction process, including pre-processing, separation, and extraction. The pre-processing step uses the guide image filter to remove various kinds of noise from the image. Improved binomial thresholding (IBT) separates the text from the image. Then the tabular text is recognized and extracted from CDI using deep neural network (DNN). In this work, weights of DNN layers are optimized by the Harris Hawk optimization algorithm (HHOA). Obtained text and associated information can be used in many ways, including replicating the document in digital format, information retrieval, and text summarization. The proposed process is applied comprehensively to UNLV, TableBank, and ICDAR 2013 image datasets. The complete procedure is implemented in Python, and precision metrics performance is verified.

A review of digital watermarking techniques: Current trends, challenges and opportunities

The rapid progress in the transfer of information and its availability are the reasons behind the widespread use of the Internet. Document images are the most complicated and challenging category among various forms of digital data to which watermarking can be applied for its security and authentication. Watermarking of the document image is a very difficult task, since these images have very limited redundancy. Because of this factor there has been very little research in this area. A comprehensive research should be carried out to ensure the effective assessment, review as well as implementation of document image watermarking techniques. This paper reviews the existing document image watermarking by considering different evaluation parameters. Based on this review, a variety of challenges and directions have been prescribed for the development of effective watermarking techniques for document images.

Enhancing Document Image Retrieval in Education: Leveraging Ensemble-Based Document Image Retrieval Systems for Improved Precision

Document image retrieval (DIR) systems simplify access to digital data within printed documents by capturing images. These systems act as bridges between print and digital realms, with demand in organizations handling both formats. In education, students use DIR to access online materials, clarify topics, and find solutions in printed textbooks by photographing content with their phones. DIR excels in handling complex figures and formulas. We propose using ensembles of DIR systems instead of single-feature models to enhance DIR’s efficacy. We introduce “Vote-Based DIR” and “The Strong Decision-Based DIR”. These ensembles combine various techniques, like optical code reading, spatial analysis, and image features, improving document retrieval. Our study, using a dataset of university exam preparation materials, shows that ensemble DIR systems outperform individual ones, promising better accuracy and efficiency in digitizing printed content, which is especially beneficial in education.

Table structure recognition using black widow based mutual exclusion and RESNET attention model

Tables are commonly used for effective and compact representation of relational information across the data in diverse document classes like scientific papers, financial statements, newspaper articles, invoices, or product descriptions. However, table structure detection is a relatively simple process for humans, but recognizing precise table structure is still a computer vision challenge. Further, innumerable possible table layouts increase the risk of automatic topic modeling and understanding the capability of each table from the generic document. This paper develops the framework to recognize the table structure from the Compound Document Image(CDI). Initially, the bilateral filter is designed for image transformation, enhancing CDI quality. An improved binarization-Sauvola algorithm (IBSA) is proposed to degrade the tables with uneven illumination, low contrast, and uniform background. The morphological Thinning method extracts the line from the table. The masking approach extracts the row and column from the table. Finally, the ResNet Attention model optimized over Black Widow optimization-based mutual exclusion (BWME) is developed to recognize the table structure from the document images. The UNLV, TableBank, and ICDAR-2013 table competition datasets are used to evaluate the proposed framework’s performance. Precision and accuracy are the metrics considered for evaluating the proposed framework performance. From the experimental results, the proposed framework achieved a precision value of 96.62 and the accuracy value of 94.34, which shows the effectiveness of the proposed approach’s performance.

Towards Removal of Shadows Caused due to Object Interferences in Smartphone Captured Document Images using Multiple Mask Generation Technique

Shadows cause a reduction of text legibility and contrast of the overall document, leading to poor recognition rates by Optical Character Recognition (OCR). Shadows were cast during photo capturing because of object interferences due to projection angle from illumination sources and are challenging to handle by OCR. This work proposes an automated technique to detect the shadows caused by object interferences and eliminate them from document images. The proposed algorithm generates two reference masks of shadow and non-shadow regions using a global thresholding method to detect and remove shadow. Initially, the document image is pre-processed, followed by mask generation for shadow and non-shadow regions, and finally, a global thresholding strategy that removes the shadow using an image mapping technique. To evaluate the effectiveness of the proposed approach, a dataset of real-world smartphone-captured document images of about 300 is considered. The datasets encompass Mobile Dataset 3 and Mobile Dataset 4 from DIBCO datasets. Experiments are conducted to quantify the efficiency of the proposed approach using performance measures such as recognition rate by OCR, presentation score, legibility score, skew, and overall document quality score. From the results, the proposed approach collectively produces an average recognition rate of 87.15% and 81.50% towards word-level recognition of Mobile Datasets 3 and 4. The text quality of the document image post shadow removal procedure achieves a higher performance score with an average presentation score of 0.35, legibility score of 0.78, skew as 73.45 degrees from the horizontal axis, and overall document quality score above six towards datasets considered. Based on qualitative analysis, comparative analysis proves that the proposed is simple yet more proficient than existing methods.

Comparative Analysis of HSV and UV-Based Approaches Denoising Degraded Document Images

The restoration of degraded document images is a critical aspect of document image processing. In this study, we present a rigorous comparative analysis of two denoising techniques applied to degraded document images within the HSV and UV color spaces. The primary objective is to quantitatively assess the efficacy of these methods in enhancing the quality of degraded document images. Experiments were conducted using an extensive dataset of degraded document images, and a comprehensive evaluation was performed employing established metrics, including peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and mean square error (MSE). Results unequivocally demonstrate the superiority of the UV-based approach over the HSV-based method across all evaluation criteria. Specifically, the UV-based approach consistently outperforms the HSV-based approach in PSNR, SSIM, and MSE assessments. Qualitative analysis further underscores the remarkable effectiveness of both techniques in elevating the quality of degraded document images. This study addresses a pressing problem in document image processing by delivering invaluable insights into the comparative efficacy of these denoising techniques. It offers a useful tool for academics and professionals who want to improve the clarity and legibility of deteriorated document images.

Pashto script and graphics detection in camera captured Pashto document images using deep learning model.

Layout analysis is the main component of a typical Document Image Analysis (DIA) system and plays an important role in pre-processing. However, regarding the Pashto language, the document images have not been explored so far. This research, for the first time, examines Pashto text along with graphics and proposes a deep learning-based classifier that can detect Pashto text and graphics per document. Another notable contribution of this research is the creation of a real dataset, which contains more than 1,000 images of the Pashto documents captured by a camera. For this dataset, we applied the convolution neural network (CNN) following a deep learning technique. Our intended method is based on the development of the advanced and classical variant of Faster R-CNN called Single-Shot Detector (SSD). The evaluation was performed by examining the 300 images from the test set. Through this way, we achieved a mean average precision (mAP) of 84.90%.

Graph attention-driven document image classification through DualTune learning

Document image classification is a challenging task due to the complexity of information contained within documents, including text, images, and their spatial arrangement. Deep learning has become a pivotal tool for extracting and learning complex patterns. However, conventional methods often grapple with integrating different data modalities and minimizing redundancy, leading to a need for more advanced and efficient deep learning strategies. This study presents a new approach to document image classification, named graph attention-driven with dual tune learning (GAD-DTL). GAD-DTL employs dual-tune learning and graph attention networks. The methodology creates semantic region embedding within document images, which incorporate both textual and spatial data. A key feature of this approach is the adaptive fusion layer, which integrates different modalities and uses a graph attention layer to capture context within each region. To minimize redundancy in learned features, we implement two distinct learning techniques, relational and non-relational learning. This approach enhances document image classification by ensuring invariant representation and minimal redundancy in features.

The Reality of High Performing Deep Learning Models: A Case Study on Document Image Classification

The Reality of High Performing Deep Learning Models: A Case Study on Document Image Classification