The multimedia communication and image-sharing platform development has increased exponentially, which enhances the threat of unauthorized entry, degradation through compression, and the manipulation of the visual data. The classical watermarking methods that are based on fixed-domain transforms are likely DCT and DWT are not always adaptive to different compression rates, and do not have high robustness to adversarial environments. In addition, methods that are available are limited by low levels of imperceptibility or computation, which hinder real time or large-scale application. In order to overcome such limitations, a new Quantum-Optimized Hierarchical Chunk Encoding (QHCE) model of perceptually adaptive and compression-sensitive image watermarking is proposed. The offered approach incorporates the entropy-based quadtree partitioning of the image into chunks, saliency-based region selection, and transform-domain embedding of watermark with the use of multi-layers DWT. Quantum Genetic Algorithm (QGA) is used to find the optimal location, band and strength of the embedding parameter that results in a good trade-off between the robustness and the visual fidelity. The model was trained with Python 3.10 and PyWavelets and Qiskit and tested on Kodak image dataset at different JPEG compression ratios. Experimental results demonstrate a significant improvement over baseline methods, with an average PSNR of 57.8 dB, SSIM of 0.997, 0.00% BER, and 100% extraction accuracy at QF ≥ 70. The QGA-based optimization also achieved a 19% increase in payload capacity and a 25% reduction in runtime compared to non-optimized baselines. Integrity verification reached 99.95% accuracy using SHA-256 and Hamming distance analysis. Integrating quantum-inspired optimization within a perceptually driven encoding pipeline offers a scalable, secure, and highly robust solution for next-generation visual data protection. This provides a compelling direction for future research in quantum-secure multimedia systems and watermark resilience.

BackgroundDeep learning models have shown strong potential for automated fracture detection in medical images. However, their robustness under varying image quality remains uncertain, particularly for small and subtle fractures, such as scaphoid fractures. Understanding how different types of image perturbations affect model performance is crucial for ensuring reliable deployment in clinical practice.ObjectiveThis study aimed to evaluate the robustness of a deep learning model trained to detect scaphoid fractures in radiographs when exposed to various image perturbations. We sought to identify which perturbations most strongly impact performance and to explore strategies to mitigate performance degradation.MethodsRadiographic datasets were systematically modified by applying Gaussian noise, blurring, JPEG compression, contrast-limited adaptive histogram equalization, resizing, and geometric offsets. Model accuracy was evaluated across different perturbation types and levels. Image quality was quantified using peak signal-to-noise ratio and structural similarity index measure to assess correlations between degradation and model performance.ResultsModel accuracy declined with increasing perturbation severity, but the extent varied across perturbation types. Gaussian blur caused the most substantial performance drop, whereas contrast-limited adaptive histogram equalization increased the false-negative rate. The model demonstrated higher resilience to color perturbations than to grayscale degradations. A strong linear correlation was found between peak signal-to-noise ratio–structural similarity index measure and accuracy, suggesting that better image quality led to improved detection. Geometric offsets and pixel value rescaling had minimal influence, whereas resolution was the dominant factor affecting performance.ConclusionsThe findings indicate that image quality, especially resolution and blurring, substantially influences the robustness of deep learning–based fracture detection models. Ensuring adequate image resolution and quality control can enhance diagnostic reliability. These results provide valuable insights for designing more accurate and resilient medical imaging models under real-world variability.

Abstract In the era of intelligent healthcare, medical images play a crucial role in remote diagnosis, disease detection, cloud storage, and data sharing. However, they are vulnerable to security threats such as data tampering, copyright disputes, and privacy breaches. Traditional digital watermarking algorithms struggle to balance imperceptibility and robustness, making them insufficient to meet the security requirements of medical image protection. Motivated by these challenges, this paper proposes an efficient, robust, secure digital watermarking scheme for medical images. The proposed method integrates Dual-Tree Complex Wavelet Transform (DTCWT), Discrete Cosine Transform (DCT), and Singular Value Decomposition (SVD) to extract image features. Particle Swarm Optimization (PSO) is then employed to adaptively adjust the watermark embedding parameters, enhancing imperceptibility and robustness. Moreover, Henon chaotic mapping is introduced to generate pseudo-random sequences for watermark encryption, further enhancing security. Experimental results show that the proposed method achieves a PSNR of 34.59 dB, ensuring good imperceptibility. Under various attacks, the extracted watermark maintains high robustness, with an average NC of 0.99 under Gaussian low-pass filtering (5 $$\times $$ × 5), 0.99 under JPEG compression (QF = 10), 0.97 under Gaussian noise, and 0.98 under rotation attacks (15 $$^{\circ }$$ ∘ ). Research shows that this method can effectively improve the secure storage and transmission capabilities of medical images, providing strong support for the security of medical image data in intelligent medical environments.

BackgroundComputer-aided detection (CAD) software provides scalable, standardized chest X-ray (CXR) interpretation, helping address the global shortage of radiologists and inter-reader variability. Printed X-ray films remain common in many low-resource settings, yet most CAD software can only process Digital Imaging and Communications in Medicine (DICOM) files. Genki software (DeepTek, India) is one of the few World Health Organization (WHO)–recommended CAD software capable of interpreting both DICOM files and photographs of printed X-ray films (Joint Photographic Experts Group [JPEG] files), but its performance using JPEG files has not been independently evaluated.MethodsWe evaluated Genki software using a test library of 1466 CXR images from adults screened for tuberculosis (TB) in Ho Chi Minh City, Viet Nam. Each participant’s TB status was determined using a composite reference standard, based on radiological findings and Xpert MTB/RIF Ultra testing. Each CXR image was blindly re-read by 10 human readers and processed by Genki software using both DICOM and JPEG files. Genki software performance was evaluated using median abnormality scores, area under the receiver operating characteristic curves (AUC), and sensitivity/specificity comparisons at different abnormality score thresholds.ResultsGenki software abnormality scores were significantly higher when using JPEG files, but this did not translate into significant differences in AUCs between the file types (DICOM AUC = 0.94 vs JPEG AUC = 0.92, p = 0.190). When abnormality score thresholds were calibrated to match average human reader sensitivity (79.0%), Genki achieved significantly higher specificity with both DICOM (95.2% vs 84.8%, p < 0.001) and JPEG (92.1% vs 84.8%, p < 0.001) files. When the software’s abnormality score thresholds were calibrated to achieve 90% sensitivity, Genki maintained high specificity with both DICOM (89.3%) and JPEG (81.1%) file types, meeting the minimum Target Product Profile (TPP) criteria for a high-sensitivity, high-specificity screening test.ConclusionsGenki software performs comparably when interpreting DICOM and JPEG files, outperforming human readers and meeting TPP criteria with both file types. This capability enhances its usability in resource-limited settings where digital infrastructure is lacking, supporting its broader deployment for TB screening. Further research is needed to assess real-world implementation feasibility and performance in diverse populations and clinical environments.Supplementary InformationThe online version contains supplementary material available at 10.1186/s44263-025-00237-8.

In the rapidly evolving field of digital security, this study aims to advance image steganography by developing and benchmarking seven deep learning architectures with a focus on imperceptibility, embedding capacity, and robustness against steganalysis. The models implemented include the residual dense network (RDN), vision transformer with adaptive attention (ViT-AA), progressive generation network (PGN), dual-stream architecture (DSA), wavelet-based hybrid network (WHN), mutual attention transformer (MAT), and efficient attention pyramid transformer (EAPT). Using the PyTorch framework and standardized datasets such as DIV2K, COCO, and ImageNet, each architecture was trained through structured preprocessing and evaluated using metrics including PSNR, SSIM, LPIPS, and statistical steganalysis resistance. Experimental results demonstrate that WHN achieved the highest visual quality (PSNR = 43.5 dB, SSIM = 0.995), while MAT and EAPT provided superior security with detection rates near random chance (0.501–0.502) and robustness against JPEG compression and noise insertion. PGN and DSA offered low-latency performance suitable for resource-constrained or mobile applications, while ViT-AA provided a balanced trade-off across imperceptibility and robustness. The findings confirm that deep learning approaches surpass traditional methods and establish new computational benchmarks for covert communication and digital forensics. These results recommend WHN, MAT, and EAPT for high-security contexts, PGN and DSA for embedded platforms, and ViT-AA as a general-purpose framework, while encouraging further research into lightweight variants for IoT and real-time deployments. Tools for data collection and experimentation included benchmark datasets and PyTorch-based implementations.

This study introduces a novel approach that integrates Gaussian Mixture Models (GMM) with MD5 hash-based verification to detect hidden messages embedded via Least Significant Bit (LSB) steganography in JPEG images. Unlike previous methods, the proposed dual-layer technique combines probabilistic modeling with data integrity verification. The model was trained and evaluated using a dataset comprising both original and stego-JPEG images. The experimental results achieved an accuracy of 78.67% and a precision of 89.15%, indicating good class separation between stego and non-stego images (AUC-ROC = 0.8659). However, the recall rate of 69.70% suggests that there is room for improvement in detecting all stego instances. Although MD5 is a hash function rather than an encryption algorithm, it effectively aids in identifying data anomalies resulting from message embedding. Overall, this lightweight approach offers a practical solution for steganalysis and can be further enhanced through the integration of hybrid deep learning techniques in future research.

Abstract The integration of Explainable AI (XAI) into healthcare promises greater transparency and interpretability of machine learning models, enabling clinicians to understand predictions and make more reliable medical decisions. Yet, the robustness of XAI methods remains uncertain, as small input perturbations can drastically change their explanations, posing critical risks in clinical settings where they may lead to misdiagnoses or inappropriate treatment. Motivated by the central role of XAI in healthcare decision-making, this paper examines its robustness in the presence of data corruption. We systematically evaluate the stability of widely used XAI techniques against both naturally occurring noise (e.g., JPEG compression) and adversarial manipulations that alter explanations without affecting model predictions. To this end, we introduce a set of evaluation metrics that capture complementary aspects of explanation stability, ranging from pixel-level consistency to spatial coherence, and propose a protocol for assessing the resilience of XAI methods across diverse perturbation sources. Our analysis spans three medical imaging datasets, various convolutional and transformer models, and ten post-hoc XAI methods, including Grad-CAM++ for convolutional networks and LibraGrad for vision transformers. We find that current XAI techniques are often unstable, even under imperceptible perturbations. For adversarial noise, a clear set of robust methods emerges, whereas for natural noise, performance varies, with some methods maintaining spatial stability and others preserving pixel-wise consistency. All results together highlight the need for multi-perspective evaluation when selecting XAI techniques in practice.

The intentional manipulation of visual data has been increasing due to the widespread use of image editing software and social media websites, challenging existing forgery detection methods. Error Level Analysis (ELA) based methods often struggle with JPEG compression, limiting their ability to detect tampering accurately. This paper proposes an adaptive compression mechanism to enhance ELA-based image forgery detection, particularly for augmented and expanded datasets. Using the CASIA V2 image forgery dataset with rotation, flipping, and scaling, ELA maps were derived and classified via a Convolutional Neural Network (CNN). The experimental results indicate that the proposed method achieved a better performance with accuracy, precision, recall, and F1-score of 96.6%, 96.8%, 96.3%, and 96.5%, respectively.

Image steganography is the process of hiding information, which can be text, image, or video inside a cover image. Recent steganography literature hasn't addressed the problem of loss of secret information during extraction and reliability. Hence, to reduce information loss and provide reliability between in the basic criteria, Herein, we proposed a hybrid method that utilizes the least significant bit (LSB) substitution, transppsition, magic matrix, key and Advance Encrytion Standard (AES) algorithm. The LSB method decreases embedding errors by implementing a new value difference algorithm. In addition, to improves the reliability between the basic criterion for image steganography we used transposition, magic matrix, key and AES. The proposed method ensures a high-quality image format in the RGB color model to conceal the hidden message within the cover image which is jpeg. The proposed hybrid method performed several experiments and these are mainly based on quality assessment metrics such as PSNR, SSIM, RMSE, NCC, etc. which showed better results. The proposed method also analyzed with different perspectives in terms of different dimensions of images and different sizes of message text which showed better results. In addition, the performance of the proposed method showed better results based on (regular and singular) steganalysis, noise, and cropping attacks. The security analyses such as key space, differential, and statistical attacks show that the proposed scheme is secure and robust against channel noise and JPEG compression.

To address the scarcity of high-quality instance segmentation data for the unique architecture of high-density East Asian cities, this paper introduces the Connected Building Landscape dataset, a vital resource for urban planning, architectural analysis, and computer vision tasks. Researchers can utilize this dataset to train and benchmark a wide range of segmentation models, enabling a deeper understanding of East Asian architectural morphology. This detailed analysis provides urban planners with more precise tools and facilitates the generation of the hazard distribution maps or cultural heritage preservation priority reports. The dataset contains 2,801 JPEG images (1024 × 768 pixels) with polygonal segmentation masks, capturing diverse architectural styles and complex spatial layouts along National Route 1 from Tokyo to Osaka. All annotations are provided in a standard JSON format for easy model integration. Baseline experiments on models like Mask R-CNN and Mask2Former validate the dataset’s quality and robustness for its intended tasks.

Image steganography deals with the hidden transmission of information whereby a secret image is embedded within a cover image in a way that the secret image cannot be easily identified. In this research, we propose a steganographic system that combines edge-aware attention mechanisms using Holistically-Nested Edge Detection (HED) within deep learning frameworks to direct adaptive data embedding operations. The system starts by extracting edge maps using HED then converting them to an attention maps. These high-resolution distance-based attention maps direct adaptive bit embedding operations within cover images by adjusting the number of hidden bits per pixel through attention strength to maintain a balance between capacity and distortion. After that, we embed the secret image within the cover image based on these attention maps. In the embedding process, we use a custom adaptive Least Significant Bits (LSBs) strategy, which follows the predicted attention map that is generated from trained encoder-decoder CNN. On other hand, we optimize the embedding process using a genetic algorithm (GA) to enhance the embedding process through adjusting the threshold values of attention map rules. In this work, we use two datasets (USC-SIPI as secret images and Boss Base as cover images) to test and train our stenographic system. We assessed the performance of our optimized deep learning model based on various performance metrics like Mean Square Error (MSE), Peak Signal to Noise Ratio (PSNR), Mean Absolute Error (MAE), Structural Similarity Index (SSIM), Image Fidelity (IF), Payload Capacity (PC), Bit per pixel (BPP), Xu-Net, Ye-Net, and RS steganalysis analysis. The experimental results show that the PSNR value of 60.72–61.20 dB emerges from 0.1 BPP. The SSIM values of 0.9995–0.9996 emerge from this level. The PSNR values reach 58.27 dB and 55.16 dB when the BPP ratio reaches 0.195 and 0.397 respectively while maintaining a textrm{BER}=0/textrm{BR}=100% throughout. The steganalyzers fail to detect the hidden data because XuNet and YeNet achieve AUC values between 0.47 and 0.53 while RS analysis produces incorrect embedding rate predictions between 0.63 and 0.65. The system maintains its robustness against salt-and-pepper noise because it achieves BER values of (textrm{BER}approx 0.005 text {at } p=0.01; approx 0.051 text {at } p=0.10) whereas cropping, additive Gaussian noise, and JPEG compression drive extraction toward random (textrm{BER}approx 0.5). Runtime is modest (per image): text {embedding} approx 0.04text {-}0.17,textrm{s}, text {extraction} approx 0.02text {-}0.11,textrm{s}. The proposed method achieves a good trade-off between capacity and imperceptibility and security at low computational cost while keeping the remaining vulnerabilities under cropping and compression attacks.

Copyright infringement becomes a significant concern for the creators, which has the rights to the original multimedia data. Image watermarking is increasingly critical in addressing challenges associated with copyright infringement and ensuring data integrity in multimedia technologies. Image watermarking is essential for safeguarding intellectual property and maintaining the authenticity of digital images. This study presents a reinforcement learning-driven embedding framework for robust image watermarking, utilizing the Integer Wavelet Transform (IWT) and Schur decomposition. The proposed embedding using reinforcement learning aims to enhance watermark imperceptibility and robustness against various attacks. The watermark embedding process involves decomposing the host image using IWT, applying Schur decomposition to selected blocks, and optimizing the embedding process through reinforcement learning. This dynamic embedding of a watermark improves robustness while maintaining imperceptibility. The experiments have been tested under various attacks, including noise attacks, compression, and filtered images. Experimental results demonstrate that the proposed method achieves a PSNR of 32.29 dB and an SSIM of 0.9828, indicating high imperceptibility. Robustness tests indicate a high NC of 0.9977 under JPEG2000 compression (5:1) and 0.8695 under cropping attacks (50%), outperforming existing methods. These findings suggest that the proposed scheme is a viable solution for practical digital copyright protection applications. The results indicate that the proposed scheme not only enhances robust copyright protection but also ensures the quality of the watermarked image.

We present a meticulously curated, long-term (1981–2024) dataset documenting maize phenology dynamics across Northeast China, the nation’s most critical commercial grain base. Derived from 61 national agrometeorological stations, it captures the timing of 10 pivotal phenological stages (sowing, emergence, three-leaf, seven-leaf, jointing, tasseling, flowering, silking, milking, maturity) and derives the durations of 4 growth period lengths (sowing-jointing, jointing-silking, silking-maturity, sowing-maturity). The dataset underwent a rigorous, multi-tiered quality control protocol, including automated checks for internal consistency and expert arbitration for ambiguous records, ensuring high integrity. Subsequent analysis employed kernel density estimation to characterize the probability distribution of phenological events and univariate linear regression to quantify decadal trends. The resulting repository is substantial, comprising 976 georeferenced diagnostic plots in JPEG format and two primary data tables in XLSX format, with a total volume of 601.04 MB. Systematically organized by province and station, this dataset serves as a foundational empirical resource for quantifying climate-driven shifts in crop development, enhancing the parameterization and validation of process-based crop models, and informing the development of optimized cultivation practices and regional climate adaptation frameworks.

This paper addresses the critical challenge of detecting content-aware image manipulations, specifically focusing on seam carving forgery. While deep learning models, particularly Convolutional Neural Networks (CNNs), have shown promise in this area, their black-box nature limits their trustworthiness in high-stakes domains like digital forensics. To address this gap, we propose and validate a framework for interpretable forgery detection, termed E-XAI (Ensemble Explainable AI). Conceptually inspired by Ensemble Learning, our framework's novelty lies not in combining predictive models, but in integrating a multi-perspective ensemble of explainability techniques. Specifically, we combine SHAP for fine-grained, pixel-level feature attribution with Grad-CAM for region-level localization to create a more robust and holistic interpretation of a single, custom-trained CNN's decisions. Our approach is validated on a purpose-built, balanced, binary-class dataset of 10,300 images. The results demonstrate high classification performance on an unseen test set, with a 95% accuracy and a 99% precision for the forged class. Furthermore, we analyze the model's robustness against JPEG compression, a common real-world perturbation. More importantly, the application of the E-XAI framework reveals how the model identifies subtle forgery artifacts, providing transparent, visual evidence for its decisions. This work contributes a robust end-to-end pipeline for interpretable image forgery detection, enhancing the trust and reliability of AI systems in information security.

Digital image watermarking is a vital tool for copyright protection and content authentication. However, most existing methods perform well only under single noise types, while real-world applications often involve composite noises with multiple distortions, leading to poor robustness. To address this issue, we propose a robust image watermarking scheme. To improve performance under combined noise conditions, a two-stage training strategy is introduced: in the first stage, noise intensity increases gradually to stabilize training; in the second stage, mixed strong noises are applied to enhance generalization against complex attacks. Specifically, a strength-balanced watermark optimization algorithm is employed during the testing stage to improve visual quality while maintaining strong robustness. Furthermore, to improve robustness against JPEG compression, we adopt a differentiable fine-grained JPEG module that accurately simulates real compression and enables gradient backpropagation during training. Experimental results demonstrate the superiority of the proposed method under various single and combined distortions. Under noise-free conditions, it achieves 0% bit error rate and 53.55 dB PSNR. Under composite distortions, our scheme maintains a low average BER of 2.40% and a PSNR of 42.70 dB.

Image steganography aims to embed secret information into a cover image in such a manner that the hidden content remains visually imperceptible while still being accurately recoverable when needed. However, traditional image steganography methods often suffer from limited robustness and are highly susceptible to common image distortions such as Gaussian noise, Poisson noise, and lossy compression. To address these limitations, this article proposes DERIS, a robust image steganography model based on invertible neural networks (INNs), which enhances resistance to image distortions through structural design. The model integrates identical denoising enhancement modules both before the discrete wavelet transform (DWT) and after the inverse discrete wavelet transform (IDWT) in the backward extraction pathway, significantly improving the quality of the extracted secret images. Furthermore, a training strategy that incorporates denoising enhancement is employed to ensure the model’s stability and reversibility under various types of image interference. Extensive experiments were conducted primarily on the DIV2K, ImageNet, and COCO datasets, using evaluation metrics including Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Learned Perceptual Image Patch Similarity (LPIPS), and Normalized Cross-correlation (NCC). Experimental results demonstrate that under Gaussian noise (σ = 10), the proposed method achieves a PSNR of 32.43 dB between cover and container images, and 30.24 dB between secret and extracted images on the DIV2K dataset, representing an improvement of 1.08 dB over Pris. Under JPEG compression (QF = 80), the method achieves PSNR values of 28.63 dB (cover-container) and 27.74 dB (secret-extracted) on the ImageNet dataset, which are 2.11 dB higher than those of Pris. Similarly, on the COCO dataset under the same attack condition, the method achieves PSNR values of 28.44 dB (cover-container) and 26.81 dB (secret-extracted), showing improvements of 0.91 dB over Pris. These results significantly outperform those of current state-of-the-art methods, demonstrating the enhanced robustness and practicality of the proposed approach.

Image steganography is often employed in information security and confidential communications, yet it typically faces challenges of imperceptibility and robustness during transmission. Meanwhile, insufficient attention has been paid to preserving the quality of the secret image after JPEG compression at the receiver, which limits the effectiveness of steganography. In this study, we propose an anti-compression attention-based diffusion pattern steganography model using GAN (ADPGAN). ADPGAN leverages dense connectivity to fuse shallow and deep image features with secret data, achieving high robustness against JPEG compression. Meanwhile, an enhanced attention module and a discriminator are employed to minimize image distortion caused by data embedding, thereby significantly improving the imperceptibility of the host image. Based on ADPGAN, we propose a novel JPEG-compression-resistant image framework that improves the quality of the recovered image by ensuring that the degradation of the reconstructed image primarily stems from sampling rather than JPEG compression. Unlike direct embedding of full-size secret images, we downsample the secret image into a secret data stream and embed it into the cover image via ADPGAN, demonstrating high distortion resistance and high-fidelity recovery of the secret image. Ablation studies validate the effectiveness of ADPGAN, achieving a 0-bit error rate (BER) under JPEG compression at a quality factor of 20, yielding an average Peak Signal-to-Noise Ratio (PSNR) of 39.70 dB for the recovered images.

Image dataset of ten durian diseases captured in real-field conditions from a family orchard in Vinh Long, Vietnam

JPEG images are widely used in multimedia transmission, such as on social media platforms, owing to their efficiency for reducing storage and transmission requirements. However, because such images may contain sensitive information, encryption is essential to ensure data privacy. Traditional image encryption schemes face challenges when applied to JPEG images, as maintaining compatibility with the JPEG structure and managing the effects of lossy compression can distort encrypted data. Existing JPEG-compatible encryption methods, such as Encryption-then-Compression (EtC) and Compression-then-Encryption (CtE), typically employ a single encryption stage, either before or after compression, and often involve trade-offs between security, storage efficiency, and visual quality. In this work, an Encryption–Compression–Encryption algorithm is presented that preserves full JPEG compatibility while combining the advantages of both EtC and CtE schemes. In the proposed method, pixel-block encryption is first applied prior to JPEG compression, followed by selective coefficient encryption after compression, in which the quantized DC coefficient differences are permuted. Experimental results indicate that the second encryption stage enhances the entropy achieved in the first stage, with both stages complementing each other in terms of resistance to attacks. The addition of this second layer does not significantly impact storage efficiency or the visual quality of the decompressed image; however, it introduces a moderate increase in computational time due to the two-stage encryption process.

Abstract More effective sorting algorithms are required due to the growing number and variety of resumes in the employment market. Identifying suitable candidates for job openings from a large pool can be both repetitive and time-consuming, potentially leading to missed opportunities or biased selections due to human error. To address this challenge, this study presents a novel CV recognition system that integrates advanced technologies: You Only Look Once for detecting key sections within CVs, Tesseract-OCR for extracting text from these sections, and a series of post-processing steps to correct any text recognition errors. Additionally, the system includes an automated data organization component that stores CV information in a database, facilitating data analytics and search operations. The system was evaluated using a public dataset of 1300 resumes in JPEG, PNG, and JPG formats, sourced from various origins and reflecting diverse formats, languages, and quality levels. Preprocessing was conducted to ensure data consistency and quality. The hyperparameters of the models were optimized using a genetic algorithm. The proposed system significantly enhances efficiency and accuracy in resume sorting, allowing HR teams to concentrate on strategic tasks and streamline the hiring process. Experimental results demonstrate the system’s effectiveness, achieving a mean average precision of 92.1%, a precision rate of 92.2%, and a recall rate of 86.0%.