Text Lines Research Articles

Cascading disasters and mental health: The February 2021 winter storm and power crisis in Texas, USA

In February 2021, the state of Texas and large parts of the US were affected by a severe cold air outbreak and winter weather event. This event resulted in large-scale power outages and cascading impacts, including limited access to potable water, multiple days without electricity, and large-scale infrastructure damage. Little is known about the mental health implications of these events, as most research has predominantly focused on the mental health effects of exposure to hurricanes, wildfires, or other natural disasters that are more commonly found in the summer months. This study aimed to analyze the crisis responses from the 2021 winter weather event in Texas using Crisis Text Line, a text-based messaging service that provides confidential crisis counseling nationwide. To date, Crisis Text Line is the largest national crisis text service, with over 8 million crisis conversations since its inception in 2013. We employed multiple analytic techniques, including segmented regression, interrupted time series, autoregressive integrated moving average (ARIMA), and difference-in-difference (DID), to investigate distinct time periods of exposure for all crisis conversations. ARIMA and DID were further utilized to examine specific crisis outcomes, including depression, stress/anxiety, and thoughts of suicide. Results found increases in total crisis conversations and for thoughts of suicide after the initial winter weather event; however, crisis outcomes varied in time. Thoughts of suicide in high-impact regions were higher across multiple time periods (e.g., 4-weeks, 3-months, 6-months, 9-months and 11-months) compared to low-impact regions and were elevated compared to pre-event time periods for 6-months and 11-months from the initial event. Total crisis volume also remained elevated for high-impact regions compared to low-impact regions up to 11-months after the beginning of the winter event. Our work highlights that cascading winter weather events, like the Texas 2021 Winter storm, negatively impacted mental health. Future research is needed across different disaster types (e.g., cascading, concurrent events) and for specific crisis outcomes (e.g., depression, suicidal ideation) to understand the optimal timing of crisis intervention post-disaster.

Nudging Patient Choice: Reducing No-Shows Using Waits Framing Messaging

Healthcare providers have long grappled with patients not showing up for their scheduled medical appointments; such no-shows lead to wasted resources and longer wait times for other patients. However, new operations research offers a promising solution to this problem. The study finds that using text message reminders that include an additional line of text indicating a potentially long wait for the next available appointment can significantly reduce no-shows by a factor of 28.6%. The intervention, called waits framing, was found to be more effective among patients who were more sensitive to wait times and when the information in the message was novel and credible. The study also uncovered the mechanism underlying the intervention. Specifically, the waits framing messages increased the perceived cost of missing an appointment, leading to a reduction in queue abandonment. This study provides insights into how behavioral science can improve service operations and help tackle challenges in healthcare delivery.

Relating the Gospels: Memory, Imitation, and the Farrer Hypothesis by Eric Eve

Reviewed by: Relating the Gospels: Memory, Imitation, and the Farrer Hypothesis by Eric Eve Daniel Glover eric eve, Relating the Gospels: Memory, Imitation, and the Farrer Hypothesis (LNTS 592; London: Bloomsbury T&T Clark, 2021). Pp. viii + 247. $115. Eric Eve's Relating the Gospels is a new study on the Synoptic Problem, arguing in favor of the Farrer Hypothesis. E. proposes that offering attention to memory and the scribal and authorial practices of imitation in the ancient world help to demonstrate the plausibility of Luke's use of Matthew and Mark. He dialogues with recent advances in memory studies and scribal compositional practices to argue in seven chapters for the plausibility of Luke's use of Matthew. The first chapter introduces the book and specifies that E. has confined this study to "comparing the viability of the Farrer hypothesis with that of its principal competitor, the Two Document Hypothesis" (3). It is somewhat disappointing that the more recent Matthean [End Page 356] Posteriority Theory did not feature prominently in this book. While no book can do everything, even a surface level engagement with this recent literature in the footnotes would have been desirable. The next two chapters supply the book's method. In chap. 2, E. establishes different models of engagement with memory studies in relation to the Synoptic Problem and how these models can help support the assumption of Marcan priority. Most interesting here is the discussion of scribal composition. E. offers evidence that scribes had a profound capacity for memory and the memorization of texts. The literary elite valued not only rote memorization of lines of text for recitation but also creative memory, which allowed for an interaction with texts even amid their performance. The best scribes could recite passages from a given text that shared a theme, word, or character, rework passages around other and different themes, sometimes recite a narrative in reverse order, and all of these on demand. Those interested in composition would often practice the same sorts of recollecting techniques just described but in combination with other literary works in order to produce a new, creative retelling or reworking of those older sources. These sorts of practices of creative rewriting, therefore, challenge Alan Kirk's argument (Q in Matthew [LNTS 564; London: T&T Clark, 2016]) that the Gospel authors would have been restrained by the logical ordering of written materials. (E. helpfully complicates this argument by pointing out that whatever rearranging would be necessary on Luke's use of Matthew, would also be necessary on Matthew's use of Q.) To be sure, E. does not attribute to the Gospel authors the same level of ability as those elite scribes, but he takes for granted that they would have had access to some of these same practices. In chap. 3, E. places this vision of scribal practice into conversation with recent work on ancient paideia. E. suggests that, while progymnastic exercises of reordering and rearrangement cannot account for every instance of Luke's use of Matthew, they do help the reader to anticipate the types of changes Luke would have made to Matthew, especially once one considers Luke as engaging in typical scribal practices of creative memory. The next three chapters wade "into the weeds" as close, comparative, and exegetical studies of the double and triple tradition. In chap. 4, E. discusses and evaluates significant similarities between Matthew and Luke, especially in the sayings material, but also between the first few chapters of Matthew and Luke. E. himself lists eight thematic and verbal parallels that are difficult to explain without some sort of literary dependence. He finds much the same sort of information in their narratives of John the Baptist, the temptation, and the resurrection. Luke's scribal creativity manifests throughout his Gospel, but E. observes the significance of Luke's imitative rewriting of the narrative of Samuel's infancy narrative in the composition of Jesus's birth and early years. There we find an analogue for the kind of creative alterations to Luke's Septuagintal source that we may observe in Luke's Matthean source under the Farrer Hypothesis. Eve's last two chapters deal with problem passages in Luke, including...

Line-Level Layout Recognition of Historical Documents with Background Knowledge

Digitization and transcription of historic documents offer new research opportunities for humanists and are the topics of many edition projects. However, manual work is still required for the main phases of layout recognition and the subsequent optical character recognition (OCR) of early printed documents. This paper describes and evaluates how deep learning approaches recognize text lines and can be extended to layout recognition using background knowledge. The evaluation was performed on five corpora of early prints from the 15th and 16th Centuries, representing a variety of layout features. While the main text with standard layouts could be recognized in the correct reading order with a precision and recall of up to 99.9%, also complex layouts were recognized at a rate as high as 90% by using background knowledge, the full potential of which was revealed if many pages of the same source were transcribed.

Omit needless words: Sentence length perception.

Short sentences improve readability. Short sentences also promote social justice through accessibility and inclusiveness. Despite this, much remains unknown about sentence length perception-an important factor in producing readable writing. Accordingly, we conducted a psychophysical study using procedures from Signal Detection Theory to examine sentence length perception in naive adults. Participants viewed real-world full-page text samples and judged whether a bolded target sentence contained more or fewer than 17 words. The experiment yielded four findings. First, naïve adults perceived sentence length in real-world text samples quickly (median = 300-400 ms) and precisely (median = ~90% correct). Second, flipping real-world text samples upside-down generated no reaction-time cost and nearly no loss in the precision of sentence length perception. This differs from the large inversion effects that characterize other highly practiced, real-world perceptual tasks involving canonically oriented stimuli, most notably face perception and reading. Third, participants significantly underestimated the length of mirror-reversed sentences-but not upside-down, nor standard sentences. This finding parallels participants' familiarity with commonly occurring left-justified right-ragged text, and suggests a novel demonstration of left-lateralized anchoring in scene syntax. Fourth, error patterns demonstrated that participants achieved their high speed, high precision sentence-length judgments by heuristically counting text lines, not by explicitly counting words. This suggests practical advice for writing instructors to offer students. When copy editing, students can quickly and precisely identify their long sentences via a line-counting heuristic, e.g., "a 17-word sentence spans about 1.5 text lines". Students can subsequently improve a long sentence's readability and inclusiveness by omitting needless words.

The Use of Crisis Services Following the Mass School Shooting in Uvalde, Texas: Quasi-Experimental Event Study.

Mass shootings result in widespread psychological trauma for survivors and members of the affected community. However, less is known about the broader effects of indirect exposure (eg, media) to mass shootings. Crisis lines offer a unique opportunity to examine real-time data on the widespread psychological effects of mass shootings. Crisis Text Line is a not-for-profit company that provides 24/7 confidential SMS text message-based mental health support and crisis intervention service. This study examines changes in the volume and composition of firearm-related conversations at Crisis Text Line before and after the mass school shooting at Robb Elementary School on May 24, 2022, in Uvalde, Texas. A quasi-experimental event study design was used to compare the actual volume of firearm-related conversations received by Crisis Text Line post shooting to forecasted firearm conversation volume under the counterfactual scenario that a shooting had not occurred. Conversations related to firearms were identified among all conversations using keyword searches. Firearm conversation volume was predicted using a seasonal autoregressive integrated moving average model trained on the 3 months of data leading up to the shooting. Additionally, proportions of issue tags (topics coded post conversation by volunteer crisis counselors at Crisis Text Line after the exchange) were compared in the 4 days before (n=251) and after (n=417) the shooting to assess changes in conversation characteristics. The 4-day window was chosen to reflect the number of days conversation volume remained above forecasted levels. There was a significant increase in the number of conversations mentioning firearms following the shooting, with the largest spike (compared to forecasted numbers) occurring the day after the shooting (n=159) on May 25, 2022. By May 28, the volume reverted to within the 95% CI of the forecasted volume (n=77). Within firearm conversations, "grief" issue tags showed a significant increase in proportion in the week following the shooting, while "isolation/loneliness," "relationships," and "suicide" issue tags showed a significant decrease in proportions the week following the shooting. The results suggest that the Uvalde school shooting may have contributed to an increase in demand for crisis services, above what would be expected given historical trends. Additionally, we found that these firearm-related crises conversations immediately post event are more likely to be related to grief and less likely to be related to suicide, loneliness, and relationships. Our findings provide some of the first data showing the real-time repercussions for the broader population exposed to school shooting events. This work adds to a growing evidence base documenting and measuring the rippling effects of mass shootings outside of those directly impacted.

Age detection from handwriting using different feature classification models

Digitized handwritten documents have been used for various purposes, including age detection, a crucial area of research in fields like forensic investigation and medical diagnosis. Automated age recognition is deemed to be a difficult task, due to the great degree of similarity and overlap across people’s handwriting. Consequently, the efficiency of the classification system is determined by extracting pertinent features from handwritten documents. This research proposes a set of age-related features suggested by a graphologist to classify handwritten documents into two age groups: youth adult and mature adult. The extracted features are: slant irregularity (SI), pen pressure irregularity (PPI), text lines irregularity (TLI) and the percentage of black and white pixels (PWB). Support Vector Machines (SVM) and Neural Network (NN) classifiers have been used to train, validate and test the proposed approach using two different datasets: the FSHS and the Khatt datasets. When applied to the FSHS dtaset using SVM and NN approaches, the proposed method resulted in a classification accuracy of 71% and 63.5%, respectively. Meanwhile, when applied to the Khatt dataset, our method outperformed state-of-the-art methods with a classification accuracy of 65.2% and 67% utilising SVM and NN classifiers, respectively. These are the best rates available right now in this field.

Koronavírus na Slovensku na rozhraní východu a západu

The aim of the presented article is to collect selected thematic-motivic lines of narrative texts on geopolitical specifics of coronavirus Covid-19 discurs in Slovakia, their classification and critical reasoning - pointing out the prejudices, stereotypes, mental patterns and cultural backround present in the Slovak society. The presented intention is based on collection of narrative texts from public discussion forums available on Slovak newspaper webpages. The study intends to be a critical survey of narrative motives present in discussion forums which included geopolitical questions – coronavirus as an element in the Eastern or Western axiomatic orientation of the Slovak society. The study aims to collect and classify motives in the mentioned topic to show mental stereotypes in the Slovak society and make way for further analysis or research on the cultural and axiomatic background of the Slovak society at the geopolitcal border between East and West.

Examining Hurricane Ida's Impact on Mental Health: Results From a Quasi-Experimental Analysis.

Limited research has evaluated the mental health effects during compounding disasters (e.g., a hurricane occurring during a pandemic), and few studies have examined post-disaster mental health with alternative data sources like crisis text lines. This study examined changes in crisis help-seeking for individuals in Louisiana, USA, before and after Hurricane Ida (2021), a storm that co-occurred during the COVID-19 pandemic. An interrupted time series analysis and difference-in-difference analysis for single and multiple group comparisons were used to examine pre-and post-changes in crisis text volume (i.e., any crisis text, substance use, thoughts of suicide, stress/anxiety, and bereavement) among help-seeking individuals in communities that received US Federal Emergency Management Agency individual and public assistance following a presidential disaster declaration. Results showed a significant increase in crisis texts for any reason, thoughts of suicide, stress/anxiety, and bereavement in the four-week, three-month, and four-month post-impact period. Findings highlight the need for more mental health support for residents directly impacted by disasters like Hurricane Ida.

Large-scale genealogical information extraction from handwritten Quebec parish records

This paper presents a complete workflow designed for extracting information from Quebec handwritten parish registers. The acts in these documents contain individual and family information highly valuable for genetic, demographic and social studies of the Quebec population. From an image of parish records, our workflow is able to identify the acts and extract personal information. The workflow is divided into successive steps: page classification, text line detection, handwritten text recognition, named entity recognition and act detection and classification. For all these steps, different machine learning models are compared. Once the information is extracted, validation rules designed by experts are then applied to standardize the extracted information and ensure its consistency with the type of act (birth, marriage and death). This validation step is able to reject records that are considered invalid or merged. The full workflow has been used to process over two million pages of Quebec parish registers from the 19–20th centuries. On a sample comprising 65% of registers, 3.2 million acts were recognized. Verification of the birth and death acts from this sample shows that 74% of them are considered complete and valid. These records will be integrated into the BALSAC database and linked together to recreate family and genealogical relations at large scale.

Crisis Response and Suicidal Behaviors of Essential Workers and Children of Essential Workers During the COVID-19 Pandemic.

The COVID-19 pandemic has put unprecedented stress on essential workers and their children. Limited cross-sectional research has found increases in mental health conditions from workload, reduced income, and isolation among essential workers. Less research has been conducted on children of essential workers. We examined trends in the crisis response of essential workers and their children from April 2020 through August 2021. We investigated the impact during 3 periods of the pandemic on workers and their children using anonymized data from the Crisis Text Line on crisis help-seeking texts for thoughts of suicide or active suicidal ideation (desire, intent, capability, time frame), abuse (emotional, physical, sexual, unspecified), anxiety/stress, grief, depression, isolation, bullying, eating or body image, gender/sexual identity, self-harm, and substance use. We used generalized estimating equations to study the longitudinal change in crisis response across the later stages of the pandemic using adjusted odds ratios (aORs) for worker status and crisis outcomes. Results demonstrated higher odds of crisis outcomes for thoughts of suicide (aOR = 1.06; 95% CI, 1.00-1.12) and suicide capability (aOR = 1.14; 95% CI, 1.02-1.27) among essential workers than among nonessential workers. Children of essential workers had higher odds of substance use than children of nonessential workers (aOR = 1.33; 95% CI, 1.08-1.65), particularly for Indigenous American children (aOR = 2.76; 95% CI, 1.35-5.36). Essential workers (aOR = 1.17; 95% CI, 1.07-1.27) and their children (aOR = 1.18; 95% CI, 1.07-1.30) had higher odds of grief than nonessential workers and their children. Essential workers and their children had elevated crisis outcomes. Immediate and low-cost psychologically supportive interventions are needed to mitigate the mental health impacts of the COVID-19 pandemic on these populations.

Irregular Scene Text Detection Based on a Graph Convolutional Network.

Detecting irregular or arbitrary shape text in natural scene images is a challenging task that has recently attracted considerable attention from research communities. However, limited by the CNN receptive field, these methods cannot directly capture relations between distant component regions by local convolutional operators. In this paper, we propose a novel method that can effectively and robustly detect irregular text in natural scene images. First, we employ a fully convolutional network architecture based on VGG16_BN to generate text components via the estimated character center points, which can ensure a high text component detection recall rate and fewer noncharacter text components. Second, text line grouping is treated as a problem of inferring the adjacency relations of text components with a graph convolution network (GCN). Finally, to evaluate our algorithm, we compare it with other existing algorithms by performing experiments on three public datasets: ICDAR2013, CTW-1500 and MSRA-TD500. The results show that the proposed method handles irregular scene text well and that it achieves promising results on these three public datasets.

End-to-End Transcript Alignment of 17th Century Manuscripts: The Case of Moccia Code.

The growth of digital libraries has yielded a large number of handwritten historical documents in the form of images, often accompanied by a digital transcription of the content. The ability to track the position of the words of the digital transcription in the images can be important both for the study of the document by humanities scholars and for further automatic processing. We propose a learning-free method for automatically aligning the transcription to the document image. The method receives as input the digital image of the document and the transcription of its content and aims at linking the transcription to the corresponding images within the page at the word level. The method comprises two main original contributions: a line-level segmentation algorithm capable of detecting text lines with curved baseline, and a text-to-image alignment algorithm capable of dealing with under- and over-segmentation errors at the word level. Experiments on pages from a 17th-century Italian manuscript have demonstrated that the line segmentation method allows one to segment 92% of the text line correctly. They also demonstrated that it achieves a correct alignment accuracy greater than 68%. Moreover, the performance achieved on widely used data sets compare favourably with the state of the art.

Minors' Experiences Accessing Confidential Contraception in Texas

PurposeTexas is one of 24 states that restricts minors' ability to obtain contraception without parental consent, unless they access confidential services at federally funded Title X clinics. This study explores Texas minors' reasons for and experiences seeking confidential contraception. MethodsBetween September 2020 and June 2021, we conducted in-depth phone interviews with 28 minors aged 15–17 years. Participants were recruited via the text line and Instagram account of an organization that helps young people navigate Texas' parental consent laws. Interview transcripts were coded and analyzed using inductive and deductive codes in our thematic analysis. ResultsParticipants wanted to be proactive about preventing pregnancy by using more effective contraceptive methods but faced resistance from adults when they initiated conversations about sex and contraception or tried to obtain consent. In the absence of adult support, they turned to online and social media resources for information about types of contraception but encountered challenges finding accurate information about where to obtain methods in Texas without a parent. Only 10 participants were able to attend an appointment for contraception. Parents' increased monitoring of minors' activities during the COVID-19 pandemic, combined with transportation and appointment-scheduling barriers, made it difficult for minors to attend in-person visits, particularly if clinics were farther away. DiscussionMinors in Texas faced a range of barriers to finding accurate information and obtaining confidential contraceptive services, which were exacerbated by the COVID-19 pandemic. Expanding options for accessible confidential contraception, along with repealing parental consent laws, would better support minors' reproductive autonomy.

Sentiment Analysis On Movie Streaming Services On Twitter Using LDA and SVM Methods

Many online platforms for movie streaming have emerged nowadays in Indonesia with the development of technology. Streaming platforms that are widely used by the public are Netflix, Disney+, HBO Go, WeTV, Vidio. A lot of comparisons between streaming platforms have become a topic of discussion on social media. The opinions expressed by users of the streaming platform greatly affect other users who want to watch the movie. The transmission of opinions from users becomes positive and negative comments that are received by the public and spread widely on social media. Opinions that contain both positive and negative comments become a problem and can affect other users who want to watch movies through streaming platforms. The opinions of users about the streaming platform can be analyzed through sentiment analysis. At the stage of data collection using scrapy framework tools with Python, as much as 1,500 lines of text are preprocessed. LDA method to present topics that contain representative words. Classification using the Support Vector Machine (SVM) gets more positive comments than negative comments. The test results obtained an accuracy of 0.82, recall of 0.8, F1 score of 0.79, and precision of 0.8.

End-to-End Handwritten Paragraph Text Recognition Using a Vertical Attention Network.

Unconstrained handwritten text recognition remains challenging for computer vision systems. Paragraph text recognition is traditionally achieved by two models: the first one for line segmentation and the second one for text line recognition. We propose a unified end-to-end model using hybrid attention to tackle this task. This model is designed to iteratively process a paragraph image line by line. It can be split into three modules. An encoder generates feature maps from the whole paragraph image. Then, an attention module recurrently generates a vertical weighted mask enabling to focus on the current text line features. This way, it performs a kind of implicit line segmentation. For each text line features, a decoder module recognizes the character sequence associated, leading to the recognition of a whole paragraph. We achieve state-of-the-art character error rate at paragraph level on three popular datasets: 1.91% for RIMES, 4.45% for IAM and 3.59% for READ 2016. Our code and trained model weights are available at https://github.com/FactoDeepLearning/VerticalAttentionOCR.

Character-Aware Sampling and Rectification for Scene Text Recognition

Curved scene text recognition is a challenging task in multimedia society due to large shape and texture variance. Previous methods address this challenge by extracting and rectifying text line with equidistantly sampling, which ignore character level information and lead to distorted characters. To address this issue, this paper proposes a Character-Aware Sampling and Rectification (CASR) module, which rectifies irregular text instance according to the location and orientation information of each individual character. Specifically, CASR regards each character as a basic unit and predicts the character-level attributes for sampling and rectification. Our module not only exploits detailed character information to obtain better rectification of text line, but also employs character-level supervision in training process. In addition, CASR provides a plug-and-play module which can be easily incorporated to existing text recognition pipeline. Extensive experiments on several benchmarks demonstrate that our method obtains more accurate rectified text instances and achieves promising performance. We will release our code and models in the future.

Text Growing on Leaf

Irregular-shaped texts bring challenges to Scene Text Detection (STD). Although existing regression-based approaches achieve comparable performances, they fail to cover some highly curved ribbon-like text lines. Inspired by morphology, we found that the leaf vein can easily cover various geometries. Specifically, lateral and thin veins are emitted to margin along main vein gradually with the leaf growth. This process can decompose a concave object into consecutive convex regions, which are easier to fit. Hence, the leaf vein is suitable for representing highly curved texts. Considering the aforementioned advantage, we design a leaf vein-based text representation method (LVT), where text contour is treated as leaf margin and represented through main, lateral, and thin veins. We further construct a detection framework based on LVT, namely LeafText. In the text reconstruction stage, LeafText simulates the leaf growth process to rebuild text contours. It grows main veins in Cartesian coordinates to locate texts roughly at first. Then, lateral and thin veins are generated along the main vein growth direction in polar coordinates. They are responsible for generating the coarse contour and refining it, respectively. Meanwhile, Multi-Oriented Smoother (MOS) is designed to smooth the main vein for ensuring reliable growth directions of lateral and thin veins. Additionally, a global incentive loss is proposed to enhance the predictions of lateral and thin veins. Ablation experiments demonstrate LVT can fit irregular-shaped texts precisely and verify the effectiveness of MOS and global incentive loss. Comparisons show that LeafText is superior to existing state-of-the-art (SOTA) methods on MSRA-TD500, CTW1500, Total-Text, and ICDAR2015 datasets.

RECOGNIZING TAMIL CHARACTERS IN PALM LEAF MANUSCRIPTS (DEEP LEARNING)

Tamil is an ancient language that has a vast collection of literature written on palm leaves and other materials. Palm leaf manuscripts have been used as a versatile medium to record information on medicine, literature, theatre, and other subjects. Despite the need for digitization and transcription, recognizing cursive characters in palm leaf manuscripts remains a challenging task. This study introduces a novel Convolutional Neural Network (CNN) technique to train the characteristics of palm leaf characters, enabling CNN to significantly classify palm leaf characters during the training phase. Preprocessing of the input image is done using morphological operations to remove noise. Connected component analysis is a technique used in image processing to identify and label the individual connected regions, or components, in a binary image. Connected component Analysis is then used to segment the palm leaf characters, with feature processing including text line spacing, spacing without obstacle, and spacing with an obstacle. Finally, the extracted cursive characters are input into the CNN technique for final classification. Experiments are conducted using collected cursive Tamil palm leaf manuscripts to validate the performance of the proposed CNN with existing deep learning techniques in terms of accuracy, precision, recall, etc.

Confidence Estimation for Object Detection in Document Images

Deep neural networks are becoming increasingly powerful and large and always require more labelled data to be trained. However, since annotating data is time-consuming, it is now necessary to develop systems that show good performance while learning on a limited amount of data. These data must be correctly chosen to obtain models that are still efficient. For this, the systems must be able to determine which data should be annotated to achieve the best results. In this paper, we propose four estimators to estimate the confidence of object detection predictions. The first two are based on Monte Carlo dropout, the third one on descriptive statistics and the last one on the detector posterior probabilities. In the active learning framework, the three first estimators show a significant improvement in performance for the detection of document physical pages and text lines compared to a random selection of images. We also show that the proposed estimator based on descriptive statistics can replace MC dropout, reducing the computational cost without compromising the performances.