Low Quality Video Research Articles

Cross-scale hierarchical spatio-temporal transformer for video enhancement

The diversity and complexity of degradations in low-quality videos pose non-trivial challenges on video enhancement to reconstruct the high-quality counterparts. Prevailing sliding window based methods represent poor performance due to the limitation of window size. Recurrent networks take advantage of long-term modeling to aggregate more information, resulting in significant performance improvements. However, most of them are trained on simple degraded data and can only tackle specific degradation. To break through the limitation, we propose a progressive alignment network, namely Cross-scale Hierarchical Spatio-Temporal Transformer (CHSTT), which leverages cross-scale tokenization to generate multi-scale visual tokens in the entire sequence to capture extensive long-range temporal dependencies. To enhance the spatial and temporal interactions, we introduce an innovative hierarchical Transformer, facilitating the computation of mutual multi-head attention across both spatial and temporal dimensions. Quantitative and qualitative assessments substantiate the superior performance of CHSTT compared to several state-of-the-art benchmarks across three distinct video enhancement tasks, including video super-resolution, video denoising, and video deblurring.

YOUTUBE AS A SOURCE OF HEALTH INFORMATION: QUALITY AND RELIABILITY ANALYSIS OF SARCOPENIA-RELATED VIDEOS

Introduction. YouTube is a prominent platform for obtaining health-related information, providing expert guidance and user-generated content. This research assessed the features and quality of YouTube videos concerning sarcopenia. Methods. On September 9, 2024, YouTube videos were viewed in incognito mode using the keywords «sarcopenia» and «loss of muscle strength.» A total of 100 videos were evaluated, with 37 satisfying the inclusion criteria following exclusions. The video quality was evaluated utilizing the Global Quality Scale (GQS) and the modified DISCERN instrument. Data regarding views, likes, comments, video duration, and sources were analyzed. Statistical analyses encompassed the Kruskal-Wallis test for group comparisons and Cohen’s kappa coefficient for inter-rater reliability. Results. Of the 37 videos examined, 48.6% (n = 18) were categorized as high-quality, 32.5% (n = 12) as intermediatequality, and 18.9% (n = 7) as low-quality. Videos of high quality were predominantly obtained from physicians (75%, n = 6), non-profit organizations (71.4%, n = 5), and government agencies (66.7%, n = 2). Low-quality videos predominantly originated from independent users (100%, n = 2) and health-related websites (36.4%, n = 4). Engagement indicators, comprising daily views, likes, and comments, showed no significant differences among quality groups, underscoring a disconnect between popularity and quality. Conclusion. YouTube is a helpful platform for distributing information on sarcopenia, provided the content is derived from reputable sources. The abundance of substandard videos highlights the danger of disinformation. Enhancing content control, promoting authoritative videos, and improving health literacy among viewers is crucial for optimizing YouTube’s capacity as a dependable health information resource.

Perceived Usefulness as a Mediating Variable on the Relationship Between Video Quality and Audience Emotional Engagement in Social Media

Facing dual challenges of inadequate dissemination and limited acceptance, this study scrutinizes why social media struggles to promote traditional culture effectively. It identifies low-quality videos on social platforms, particularly short-form content, as the main hurdle, leading to reduced impact. Investigating the relationship between short-form video quality and audience receptivity, this study categorizes quality into media richness, impact, and interactivity dimensions. Findings confirm that higher-quality videos correlate with increased emotional engagement. Employing the Technology Acceptance Theory (TAM), perceived usefulness partially mediates the link between video quality and emotional engagement. Elevating video quality not only enhances perceived usefulness but also deepens emotional resonance, ultimately amplifying cultural impact.

YouTube as asource of information on reactive arthritis: aquality analysis.

YouTube is often used by patients and healthcare professionals to obtain medical information. Reactive arthritis (ReA) is atype of inflammatory arthritis triggered by infection, usually in the genitourinary or gastrointestinal tract. However, the accuracy and quality of ReA-related information on YouTube are not fully known. This study aimed to assess the reliability and quality of YouTube videos pertaining to ReA. AYouTube search was performed on August1, 2023, using the keywords "reactive arthritis," "Reiter's disease," and "Reiter's syndrome." The number of days since upload; the number of views, likes, and comments; and the duration of videos were recorded. The modified DISCERN tool (mDISCERN) and the global quality scale (GQS) were used to evaluate the reliability and quality of the videos. Two physicians independently classified videos as low, moderate, or high quality and rated them on afive-point GQS (1 = poor quality, 5 = excellent quality). The source of videos was also noted. Of the 180videos screened, 68met the inclusion criteria. The most common topic (61, 89.7%) was "ReA overview." Among the 68videos analyzed, the main source of uploads was physicians 45(66.2%), and 66(97%) were categorized as useful. Around half of the YouTube videos about ReA were of high quality (33, 48.5%) according to the GQS. Upon comparing videos uploaded by rheumatologists, non-rheumatology healthcare professionals, and independent users, significant differences were found in mDISCERN and GQS but not in the number of views, likes, and comments or duration. Upon comparing high-, moderate-, and low-quality videos, significant differences were found in the number of views, likes, and comments; duration; and in mDISCERN and GQS. YouTube is asource of information on ReA of variable quality, with wide viewership and the potential to influence patients' knowledge and behavior. Our results showed that most YouTube videos on ReA were of high quality. Videos presented by physicians had higher quality. YouTube should consider avoiding low-quality videos by using validity scales such as mDISCERN and GQS.

Content quality and reliablity of YouTube™ videos as a source of information about good oral hygiene practices in adults.

Good oral hygiene is crucial for preventing dental caries and periodontal diseases. However, proper and regular application of oral hygiene practices requires adequate knowledge. In recent years, the internet has become one of the most popular places to find health-related information, necessitating studies that analyze the quality of the content available online. The purpose of the present study was to analyze the content quality and reliability of YouTube™ videos on the topic of adult oral hygiene practices and to guide oral health care professionals who use this platform for patient education. A YouTube™ search was performed of the most frequent search term, 'dental hygiene'. A total of 150 videos were screened, and 51 were included in the final study. The characteristics, sources, and content of the videos were analyzed using the Global Quality Score (GQS) and DISCERN reliability indices. The IBM SPSS 25 program was used for statistical analyses. Most of the included videos were uploaded by oral health care professionals (63%). GQS revealed only 17.6% of the videos were excellent quality whereas 23.5% of them were poor quality. In the content analysis, 62.7% of the videos were deemed moderately useful. Video duration, total content score, and interaction indices were all significantly higher in the useful and very useful groups compared to the slightly useful group (p = 0.020, p < 0.001, p = 0.040). GQS had a positive, low-medium statistically significant correlation with both video duration and total content scores (r = 0.235, r = 0.517; p < 0.05). DISCERN score also had a positive, low-medium statistically significant correlation with total content score (r = 0.500; p < 0.05). The study concluded that most YouTube™ videos on oral hygiene practices for adults are moderately useful. When using YouTube™ for patient education, oral health care professionals and organizations should be aware of low-quality videos and seek out accurate, useful videos. There is also a need for quality videos with expanded oral health content.

Data safety prediction using YOLOv7+G3HN for traffic roads

Pulau Pinang has introduced several measures to enhance traffic safety and promote sustainability, including the installation of CCTV systems and the implementation of smart solutions and green technology as part of the Penang 2030 vision, aligning with the Sustainable Development Goals (SDGs). However, despite these efforts, road accidents persist due to non-optimised detection models, incomplete data from manual reporting, and technological constraints in real-time video analysis and predictive modelling. This study evaluates the effectiveness of the YOLOv7+G3HN framework for vehicle detection and near-miss analysis, with a focus on the influence of video quality on detection performance. The research aims to understand how high- and low-quality video inputs affect the accuracy and computational efficiency of detection algorithms. High[1]quality videos resulted in significantly faster computation times for vehicle detection than low-quality videos, highlighting the importance of video resolution in optimising detection processes. Despite the robustness of the algorithm, with no errors detected in both video qualities, higher miss detection rates in low-quality videos suggest that lower resolution may compromise detection accuracy and the effectiveness of monitoring systems. Near-miss analysis revealed that high-quality videos had a lower probability of near-miss occurrences than low-quality videos, highlighting the importance of video resolution for detection efficacy. These findings emphasise the critical role of high-resolution video inputs in enhancing detection accuracy and reliability, advocating for their implementation to optimise vehicle detection and improve road safety. Additionally, YOLOv7+G3HN outperforms YOLOv7 in both accuracy and speed. The study concludes that the YOLOv7+G3HN framework is effective for vehicle detection and near-miss analysis, provided that video quality is considered in system design and implementation.

Collapse of the Dharahara Tower During the April 25, 2015 Nepal Earthquake: A New Interpretation Based on Video-Camera Footage

ABSTRACT Among the various important monuments impacted by the April 25, 2015 Mw7.8 Nepal earthquake, the Dharahara Tower located in Kathmandu suffered complete collapse. The collapse of the tower, a slender 60-meter-high old brick masonry structure with an almost cylindrical hollow shape, was captured by a low-quality video camera, which nonetheless helped to identify the critical moments. Initially, the tower descended more than 10 m vertically after a rupture at the base. After a few seconds, it overturned, disintegrating upon impact on the ground. At a distance of 1.3 km, an accelerometric station recorded the strong motion which was used as input of a discrete element model of the structure. Mechanical properties were taken from lab tests made on similar building materials. The Dharahara Tower had previously been analysed by other researchers using various methods. However, since they were apparently not aware of the mechanism observed in the video footage, their collapse models do not agree with this new evidence. For the initial phase of the collapse, the present model explains with great accuracy the observations of the video footage, but the final phase is dependent on various details, and modelling is not so robust.

The role of a digital twin in supporting criminal investigations - a case report about a possible abuse.

As part of a comprehensive analysis, this case report presents a possible case of child maltreatment that can serve as a basis for forensic and medical examiner investigations. This case concerns the death of an infant who was approximately two months old. During a routine examination by the pediatrician at the end of May 2021, the child was found to have a normal head circumference of 31 cm. No other abnormalities were noted. On June 19, 2021, the child died, and an autopsy revealed a head circumference of 44 cm and a subdural hematoma as the cause of death. Questions arose as to who might have abused the child and when. The only evidence was a low-quality cell phone video taken by the child's parents on June 13, 2021, six days before the child's death, in which the child could be seen lying on a pillow. It was necessary to determine whether the child in this video already had an unnatural head circumference. This study presents a novel workflow that demonstrates how to analyze and deal with low quality video to answer questions like the above. The workflow demonstrates the creation of 3D scenes from digital image and video material. These 3D scenes can be used for object measurement and to support forensic and medical investigations. In the present case, where only low quality smartphone images were available, the presented workflow was used to create a 3D scene of the child lying on the pillow. In this 3D scene, it was possible to determine the child's head circumference. These measurements support the findings of the medical examiner (dated June 24, 2021) and confirm the suspicion that possible child abuse had already taken place on June 13, 2021. The innovative approach makes it possible to identify evidence of possible abuse based on a specific point in time, in this case the child's private footage. It also demonstrates the potential of 3D scene reconstruction in complex forensic and medical scenarios.

BONES: Near-Optimal Neural-Enhanced Video Streaming

Accessing high-quality video content can be challenging due to insufficient and unstable network bandwidth. Recent advances in neural enhancement have shown promising results in improving the quality of degraded videos through deep learning. Neural-Enhanced Streaming (NES) incorporates this new approach into video streaming, allowing users to download low-quality video segments and then enhance them to obtain high-quality content without violating the playback of the video stream. We introduce BONES, an NES control algorithm that jointly manages the network and computational resources to maximize the quality of experience (QoE) of the user. BONES formulates NES as a Lyapunov optimization problem and solves it in an online manner with near-optimal performance, making it the first NES algorithm to provide a theoretical performance guarantee. Comprehensive experimental results indicate that BONES increases QoE by 5% to 20% over state-of-the-art algorithms with minimal overhead. Our code is available at https://github.com/UMass-LIDS/bones.

Can Chat-GPT assist orthopedic surgeons in evaluating the quality of rotator cuff surgery patient information videos?

BackgroundPatients and healthcare professionals extensively rely on the internet for medical information. Low-quality videos can significantly impact the patient-doctor relationship, potentially affecting consultation efficiency and decision-making process. Chat Generative Pre-Trained Transformer (ChatGPT) is an artificial intelligence application with the potential to improve medical reports, provide medical information, and supplement orthopedic knowledge acquisition. This study aimed to assess the ability of ChatGPT-4 to detect deficiencies in these videos, assuming it would be successful in identifying such deficiencies. Materials and MethodsYouTube was searched for 'rotator cuff surgery' and 'rotator cuff surgery clinic' videos. A total of 90 videos were evaluated, with 40 included in the study after exclusions. Using the Google Chrome extension '' YouTube Summary with ChatGPT & Claude'', transcripts of these videos were accessed. Two senior orthopedic surgeons and ChatGPT-4 evaluated the videos using the rotator cuff surgery YouTube score (RCSS) system and DISCERN criteria. ResultsChatGPT-4's RCSS evaluations were comparable to those of the observers in 25% of instances, and 40% for DISCERN. The interobserver agreement between human observers and ChatGPT-4 was fair (AC1: 0.575 for DISCERN and AC1: 0.516 for RCSS). Even after correcting ChatGPT-4's incorrect answers, the agreement did not change significantly. ChatGPT-4 tended to give higher scores than the observers, particularly in sections related to anatomy, surgical technique, and indications for surgery. ConclusionThe use of ChatGPT-4 as an observer in evaluating rotator cuff surgery-related videos and identifying deficiencies is not currently recommended. Future studies with trained ChatGPT models may address these deficiencies and enable ChatGPT to evaluate videos at a human observer level.

Schlieren imaging and video classification of alphabet pronunciations: exploiting phonetic flows for speech recognition and speech therapy

Speech is a highly coordinated process that requires precise control over vocal tract morphology/motion to produce intelligible sounds while simultaneously generating unique exhaled flow patterns. The schlieren imaging technique visualizes airflows with subtle density variations. It is hypothesized that speech flows captured by schlieren, when analyzed using a hybrid of convolutional neural network (CNN) and long short-term memory (LSTM) network, can recognize alphabet pronunciations, thus facilitating automatic speech recognition and speech disorder therapy. This study evaluates the feasibility of using a CNN-based video classification network to differentiate speech flows corresponding to the first four alphabets: /A/, /B/, /C/, and /D/. A schlieren optical system was developed, and the speech flows of alphabet pronunciations were recorded for two participants at an acquisition rate of 60 frames per second. A total of 640 video clips, each lasting 1 s, were utilized to train and test a hybrid CNN-LSTM network. Acoustic analyses of the recorded sounds were conducted to understand the phonetic differences among the four alphabets. The hybrid CNN-LSTM network was trained separately on four datasets of varying sizes (i.e., 20, 30, 40, 50 videos per alphabet), all achieving over 95% accuracy in classifying videos of the same participant. However, the network’s performance declined when tested on speech flows from a different participant, with accuracy dropping to around 44%, indicating significant inter-participant variability in alphabet pronunciation. Retraining the network with videos from both participants improved accuracy to 93% on the second participant. Analysis of misclassified videos indicated that factors such as low video quality and disproportional head size affected accuracy. These results highlight the potential of CNN-assisted speech recognition and speech therapy using articulation flows, although challenges remain in expanding the alphabet set and participant cohort.

BONES: Near-Optimal Neural-Enhanced Video Streaming

Accessing high-quality video content can be challenging due to insufficient and unstable network bandwidth. Recent advances in neural enhancement have shown promising results in improving the quality of degraded videos through deep learning. Neural-Enhanced Streaming (NES) incorporates this new approach into video streaming, allowing users to download low-quality video segments and then enhance them to obtain high-quality content without violating the playback of the video stream. We introduce BONES, an NES control algorithm that jointly manages the network and computational resources to maximize the quality of experience (QoE) of the user. BONES formulates NES as a Lyapunov optimization problem and solves it in an online manner with near-optimal performance, making it the first NES algorithm to provide a theoretical performance guarantee. Comprehensive experimental results indicate that BONES increases QoE by 5% to 20% over state-of-the-art algorithms with minimal overhead. Our code is available at https://github.com/UMass-LIDS/bones.

Information literacy skills of health professions students in assessing YouTube medical education content

IntroductionYouTube is a popular social media video platform used by health professions students for medical education. YouTube videos vary in quality, and students need to be able to evaluate and select high-quality videos to supplement their learning. Evaluating the quality of YouTube videos is an essential information literacy skill, and the Association of College and Research Libraries updated the framework of information literacy to include collaborative platforms such as YouTube. Research is needed to understand and explore the information literacy proficiency of students in the health professions who are using YouTube videos as learning resources.MethodsThis exploratory expert-novice study investigated the information literacy proficiency of students in evaluating the quality of medical education YouTube videos. Students (n = 89) and experts (n = 23) evaluated three preselected medical education YouTube videos of varying quality using the Medical Quality Video Evaluation Tool (MQ-VET).ResultsResults of two-way mixed repeated-measures ANOVA found experts assigned significantly lower ratings to low- and medium-quality videos than students. In other words, students were less information proficient in selecting videos due to overrating source credibility, education quality, and production quality, along with having lower expectations and finding videos more useful.DiscussionThe tendency of students to overrate lower-quality videos raises concerns about their selection of educational content outside structured learning environments. If students select videos to watch outside the classroom, they could unintentionally learn medical skills from low-quality videos. These insights suggest a need for teachers to select only high-quality videos for students and for more focused curricular strategies to enhance students’ abilities in critically assessing YouTube resources for medical education.

Low-Quality Video Target Detection Based on EEG Signal Using Eye Movement Alignment.

The target detection based on electroencephalogram (EEG) signals is a new target detection method. This method recognizes the target by decoding the specific neural response when an operator observes the target, which has important theoretical and application values. This paper focuses on the EEG detection of low-quality video targets, which breaks through the limitation of previous target detection based on EEG signals only for high-quality video targets. We first design an experimental paradigm for EEG-based low-quality video target detection and propose an epoch extraction method based on eye movement signals to solve the asynchronous problem faced by low-quality video target detection. Then, the neural representation in the process of operator recognition is analyzed based on the time domain, frequency domain, and source space domain, respectively. We design the time-frequency features based on continuous wavelet transform according to the neural representation and obtain an average decoding test accuracy of 84.56%. The research results of this paper lay the foundation for the development of a video target detection system based on EEG signals in the future.

Real-World Video Super-Resolution with a Degradation-Adaptive Model.

Video super-resolution (VSR) remains challenging for real-world applications due to complex and unknown degradations. Existing methods lack the flexibility to handle video sequences with different degradation levels, thus failing to reflect real-world scenarios. To address this problem, we propose a degradation-adaptive video super-resolution network (DAVSR) based on a bidirectional propagation network. Specifically, we adaptively employ three distinct degradation levels to process input video sequences, aiming to obtain training pairs that reflect a variety of real-world corrupted images. We also equip the network with a pre-cleaning module to reduce noise and artifacts in the low-quality video sequences prior to information propagation. Additionally, compared to previous flow-based methods, we employ an unsupervised optical flow estimator to acquire a more precise optical flow to guide inter-frame alignment. Meanwhile, while maintaining network performance, we streamline the propagation network branches and the structure of the reconstruction module of the baseline network. Experiments are conducted on datasets with diverse degradation types to validate the effectiveness of DAVSR. Our method exhibits an average improvement of 0.18 dB over a recent SOTA approach (DBVSR) in terms of the PSNR metric. Extensive experiments demonstrate the effectiveness of our network in handling real-world video sequences with different degradation levels.

Effects of Browsing Conditions and Visual Alert Design on Human Susceptibility to Deepfakes

The increasing reach of deepfakes raises practical questions about people’s ability to detect false videos online. How vulnerable are people to deepfake videos? What technologies can help improve detection? Previous experiments that measure human deepfake detection historically omit a number of conditions that can exist in typical browsing conditions. Here, we operationalized four such conditions (low prevalence, brief presentation, low video quality, and divided attention), and found in a series of online experiments that all conditions lowered detection relative to baseline, suggesting that the current literature underestimates people’s susceptibility to deepfakes. Next, we examined how AI assistance could be integrated into the human decision process. We found that a model that exposes deepfakes by amplifying artifacts increases detection rates, and also leads to higher rates of incorporating AI feedback and higher final confidence than text-based prompts. Overall, this suggests that visual indicators that cause distortions on fake videos may be effective at mitigating the impact of falsified video.

Face perception in deep learning and safety

Face perception are very useful and common in our daily life today. Not only in the unlock phone but also in some crime situation. This article will introduce several way for computer to do the face perception. In this article, I will introduce the modern majority way to do the face perception. The common way to reading face is whether separate in different part or building 3D figure for face. There are several advantage by using different way like Face net, Fece++. They could be used in different condition. After reading the image from picture, we need to train computer to do the match and do better. This is the machine learning, the best tool today is CNN or python. Even all of face perception tool could not handle the low quality picture or video, but they could have 90% accuracy in certain condition. That is the accuracy for human to do it, but computer have faster speed in face perception than people did.

Quality Assessment of YouTube Videos on Avascular Necrosis of the Femoral Head: An Analysis of Content, Reliability, and Educational Value

Aim: Videos related to avascular necrosis of the femoral head are no exception, and so the aim of this study is to evaluate the educational quality of YouTube videos on this topic. &#x0D; Material and Method: A standardized video search was performed on YouTube using the terms “avascular necrosis of the hip”, “osteonecrosis of the hip” and “avascular necrosis of the femoral head”. The top 50 videos were then analyzed, and the characteristics and content of the videos were recorded. The Journal of American Medical Association criteria, The DISCERN score, The Global Quality Score and the new YouTube ‘Avascular Necrosis of the Femoral Head Score” were all used to assess the reliability and accuracy of the videos.&#x0D; Results: The median video duration of the 50 videos was 10.85±19.17 minutes. The median number of views was 10,866 (range 221 to 278,174). According to the video content, 60% of the videos contained information about the disease, 10% were about patient experience, and the remaining 30% related to surgical technique or approach. Physicians were the primary uploader on YouTube for this topic. The rate of low-quality videos was determined as following according to the different evaluative systems: 68% according to the newly defined YouTube Avascular Necrosis of the Femoral Head Score, 60% according to the DISCERN score, and 56% according to The Global Quality Score. The Journal of the American Medical Association, The Global Quality Score, and DISCERN score were significantly correlated with video duration, while the New YouTube Avascular Necrosis of the Femoral Head Score was significantly correlated with video duration, time since upload, number of views, and like rate.&#x0D; Conclusion: Most of the popular YouTube videos about avascular necrosis are of a low quality.

YouTube as a Source of Information on Public Health Ethics.

Public health ethics (PHE) is a dynamic area within bioethics that addresses the complex moral implications of public health measures in the face of growing health threats. YouTube is a powerful and widely used platform for disseminating health-related information. The primary objective of this study is to assess videos related to PHE on YouTube. The aim is to gauge the extent of misinformation in collecting PHE videos on the platform. On October 25, 2023, a thorough investigation on YouTube was undertaken, employing pre-determined search phrases involving 'public health,' 'healthcare,' 'health services administration,' and 'health policy and ethics.' The research encompassed a total of 137 videos that were selected according to strict inclusion and exclusion criteria. The videos were evaluated using the Global Quality Scale to measure quality and the modified DISCERN tool to evaluate reliability. The researchers identified video sources and compared several video attributes across different quality groups. A total of 137 videos were analyzed, and 65 (47.45%) were classified as high quality, 52 (37.23%) as moderate quality, and 21 (15.32%) as low quality. In high-quality videos, academic, government, physician, and university-hospital sources predominated, whereas Internet users and news sources were connected with low-quality videos. Significant differences in DISCERN score, per day views, likes, and comments were seen across the quality groups (P = 0.001 for views per day and P = 0.001 for other characteristics). According to the findings, low-quality videos had higher median values for daily views, likes, and comments. Although nearly half of the videos were high-quality, low-quality videos attracted greater attention. Critical contributors to high-quality videos included academic, government, physician, and university-hospital sources. The findings highlight the importance of quality control methods on social media platforms and strategies to direct users to trustworthy health information. Authors should prioritize appropriate citations and evaluate YouTube and other comparable platforms for potential promotional low-quality information.

Self-Information Forgery Mining for Face Forgery Detection

Abstract In the face of rapid advances in face forgery technology, effective detection methods have become crucial to maintain the authenticity of digital media. Deep learning technology has provided new strategies for recognizing and preventing face forgery in recent years. In this study, a new face forgery detection technique is proposed by utilizing self-information theory, which improves the accuracy and robustness of detection by mining forgery traces, especially in diverse forgery scenarios. The study extracts face features through an improved high-resolution network HRNet and optimizes identity information extraction by combining facial reenactment techniques to detect forged faces efficiently. Experiments have been conducted on several mainstream forged face datasets, and the method presented in this paper can effectively improve the detection performance with an average accuracy of 74.75% on C40 recompressed images. Comparison experiments show that this research method’s frame-level and video-level detection accuracy on the Celeb-DF dataset are 0.9846 and 0.9985, respectively, which are higher than those of existing techniques. Cross-library tests validate the method’s generalization performance, and the AUC metric remains at 0.7305 even in low-quality video environments, which shows good resistance to environmental interference. This study proposes a self-information forgery mining technique that enhances forgery detection accuracy while demonstrating superior generalization ability.