PWPatrick Shen-Pei Wang

Patrick Shen-Pei Wang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Patrick Shen-Pei Wang has a total of 39 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Trustworthiness and Authenticity in the Age of AI: Evidence-Oriented and Reliability-Aware Framework for Image Forgery Detection

    by Md Shopon, Patrick S. P. Wang, Marina Gavrilova

    The increasing accessibility of advanced image editing and generative models has strengthened the need for reliable passive image forgery detection and localization. This paper reviews representative methods for identifying manipulated images and regions without access to the original image or embedded authentication information. We introduce an evidence- and reasoning-oriented taxonomy that organizes existing approaches into five categories: acquisition consistency and noise, compression and spectral consistency, spatial correspondence and structural reasoning, multi-evidence and contextual reasoning, and generative-editing and foundation-model reasoning. The review further examines commonly used datasets, evaluation practices, and robustness settings, showing how annotation quality, manipulation realism, post-processing, and generator diversity influence reported performance. Drawing on the limitations identified across the literature, we outline the Multi-Evidence Reliability-Aware Forgery Analysis framework as a conceptual reference framework for condition-aware evidence extraction, reliability-aware fusion, uncertainty estimation, and evidence-grounded reporting. The analysis indicates that future progress depends not only on stronger architectures, but also on transferable forensic representations, realistic evaluation, calibrated uncertainty, and reliable operation under unseen manipulations and distribution conditions.

  • LiDAR: Liquid Delta-Aware Reporting Interpretable for Longitudinal Chest X-Rays

    by Anushka, Anshuman Shastri, Patrick S P Wang, Kc Santosh

    Longitudinal (i.e. follow up over time) chest X-ray (CXR) reading is inherently a comparison task — clinicians do not simply catalogue findings in a single film but instead characterize what has appeared, what has cleared, and what persists relative to the most recent prior study. Existing automated CXR systems largely ignore this workflow, treating each examination in isolation and discarding the clinically informative time gap between visits. We introduce LiDAR (Liquid Delta-Aware Reporting), a structured delta-reporting framework that directly encodes the inter-study interval as a governing signal rather than a passive feature. Patient studies are organized chronologically into prior-to-current pairs, and change targets are defined in three interpretable categories aligned with radiological language: NEW (absent before, present now), RESOLVED (present before, absent now), and STABLE PRESENT (present at both timepoints). A Liquid Neural Network cell conditions hidden-state evolution on the observed time gap, producing a mathematically transparent temporal update whose behavior is auditable without auxiliary explanation tools. Per-head thresholds calibrated on validation data decode predicted probabilities into concise, inspectable structured reports. Together, these design choices position LiDAR as an interpretable, time-aware deltare-porting system grounded in the comparison-with-priors workflow radiologists follow in practice, evaluated on PadChest.

  • Similarity-Guided Denoising Reconstruction for Unsupervised Image Captioning

    by Dongnan Yang, Lina Yang, Thomas Wu, Xichun Li, Yuan Yan Tang, Patrick Shen-Pei Wang

    Image captioning aims to generate natural and accurate textual descriptions of given images. Although significant progress has been made in image captioning models in recent years, most existing approaches heavily rely on high-quality image-text paired datasets that require expensive human annotation, thus limiting model scalability. Current unsupervised image captioning methods primarily focus on leveraging zero-shot learning capabilities of large pre-trained models (e.g. CLIP, GPT-2), yet still face persistent challenges including modality gaps, inefficient inference, and excessive noise incorporation, which constrain model accuracy and generalization capabilities. To address these limitations, we propose SGDR-Cap ( Similarity- Guided Denoising Reconstruction for Captioning), a novel unsupervised image captioning method that bridges the vision-language modality gap through a similarity-guided denoising reconstruction module. Our method leverages similarity information to guide the reconstruction of authentic text features during caption generation while simultaneously forcing the model to learn how to extract crucial image-relevant features and filter out unnecessary noise information. This enhances both coarse- and fine-grained cross-modal alignment. Furthermore, our approach jointly optimizes denoising reconstruction loss and language modeling loss, ensuring accuracy and fluency, and promoting greater diversity. Extensive evaluations on the MSCOCO and Flickr30K benchmarks demonstrate that our method achieves state-of-the-art results across all major metrics, with the most notable gain on the CIDEr score, improving from 101.1 to 104.4.

  • Unlocking the Potential of Auxiliary Captions via Dual-Branch Multi-Scale Network for Composed Image Retrieval

    by Jinhong Xu, Lina Yang, Xichun Li, Thomas Wu, Yuan Yan Tang, Patrick Shen-Pei Wang

    Composed image retrieval (CIR) aims to retrieve target images by combining a reference image with a modification text. Traditional CIR methods often struggle with feature-level multimodal fusion, leading to deviations from the original embedding space. To address this, we propose a Dual-Branch Multi-Scale Network (DMN) that integrates a combining branch and a complete text branch. To enhance the use of captions generated by advanced image captioning models for CIR, the DMN leverages an attribute-driven disentanglement layer to separate features into distinct latent factors and employs a dual-path multimodal fusion module for effective feature integration. Additionally, a multi-scale matching module incorporating both global and local matching strategies is introduced to enhance fine-grained feature discrimination. Experimental results on the FashionIQ, Shoes, and CIRR datasets demonstrate that our DMN model consistently outperforms state-of-the-art methods, achieving improvements of up to 1.43% in mean recall metrics.


  • Multi-Scale Adaptive Diffusion Feature Fusion with Causal Invariance Learning for Hyperspectral Image Classification

    by Wanxing Zha, Lina Yang, Huafu Xu, Bingzhen Wang, Danyang Chen, Yuwen Lin, Yiqun Wang, Patrick Shen-Pei Wang

    Hyperspectral image (HSI) classification is a significant research area in remote sensing with a wide range of application scenarios. Recently, numerous HSI classification methods based on convolutional neural networks (CNNs) and Transformers have demonstrated promising classification performance. However, these methods demonstrate insufficient capability in mining spectral–spatial relationships from limited HSI samples and fail to extract features pertinent to the target category. To address these challenges, we propose a multi-scale adaptive diffusion feature fusion and causal invariance learning (MSDF-CIL) framework based on diffusion models. Specifically, the framework establishes spectral–spatial distribution relationships through forward and backward diffusion processes. The forward process gradually introduces noise to the HSI input. In the backward diffusion process, our pre-trained hyperspectral denoising network extracts semantically rich multi-scale diffusion features from complex spectral–spatial relationships. A multi-scale adaptive diffusion feature fusion (MSADFF) module is designed to learn key information about each scale and fuse it to enhance the representation. In addition, a causal invariance learning (CIL) module is designed to focus on features causally related to the target class, enabling the model to eliminate spurious correlations among diffusion features. Experimental results on three public HSI datasets show that the proposed MSDF-CIL outperforms other state-of-the-art HSI classification methods, even with minimal samples. When the number of training samples in each class reaches 30, the MSDF-CIL achieves overall accuracy improvements of 4.0% on the Pavia University dataset, 2.3% on the Indian Pines dataset, and 2.3% on the Houston13 dataset, respectively.


  • STSODNet: Scale Transformer Small Object Detection Network

    by Jincheng Li, Lina Yang, Patrick Shen-Pei Wang

    Dense small object detection in complex scenes is a valuable and challenging research field. While deep learning has driven significant advancements in computer vision, traditional object detection models still struggle to achieve high accuracy in detecting small objects, particularly in large-scale aerial images. Challenges such as scale variations, occlusions, and complex backgrounds continue to hinder the effective detection of dense small objects. In this paper, we present the Scale Transformer Small Object Detection Network (STSODNet), a novel architecture designed to address these challenges. First, we conceptualize the pronounced scale variation in drone images as an anomalous disturbance and propose a multiscale feature enhancement module (MSFEM), built upon the Spatial Transformer Network, to mitigate this effect. The multiscale feature enhancement module performs learnable, multi-point magnification on regions surrounding objects based on spatial saliency, enhancing the model’s scale invariance. Second, to generate a more accurate global saliency map and heighten the model’s focus on small target regions, we introduce a refined spatial attention mechanism, termed Spatial Region Attention. This mechanism combines coarse region attention with fine spatial attention to produce a more detailed saliency map and improve long-range dependency capture. Third, to achieve more accurate spatial regression of small objects, the traditional three-layer detection head is improved by expanding its output layer, resulting in a finer and larger output while maintaining the same number of parameters. Extensive experiments on the VisDrone and SeaPerson benchmark datasets validate that STSODNet achieves superior precision and robustness, outperforming current state-of-the-art object detection methods for small object detection.


  • Analysis of Job Offers to Measure Gender Barriers through Natural Language Processing and Soft Computing Techniques

    by Cristina Puente, Ivan Sanchez-Perez, Evhenia Kolomiyets-Ludwig, Clara Palacios-Castrillo, Patrick S. P. Wang, Rafael Palacios

    Gender-biased language is still traced in job advertisements. Legal requirements to avoid direct gender-biased adjectives, and the usage of special software to detect and substitute gender-based words, scale up the issue more than solve it. The veil of discrimination on gender in job advertisements becomes more sophisticated with each succeeding level of its official and technical (including AI) prevention. This paper is mainly focused on the application of natural language processing (NLP) to detect gender-biased and discrimination of candidates by analyzing job offers posted online. NLP is an Artificial Intelligence tool that was applied in combination with Term Frequency-Inverse Document Frequency (TF-IDF) and Latent Dirichlet Allocation (LDA) to analyze the type of language used in job advertisements, detect the most relevant words used in the ads, and ultimately detect gender-bias. The main objective of this work is to provide equal access to employment opportunities from the very initial stage of the recruitment process. In addition, clustering techniques were applied to create groups based on the target public and the type of language used, providing evidence of gender-biased practices. The system was tested using a database of 2000 job ads in four different sectors: nursery, secretarial, managerial, and engineering.


  • LCSTR: Scene Text Recognition with Large Convolutional Kernels

    by Jiale Wang, Lina Yang, Jing Wang, Haoyan Yang, Lin Bai, Patrick Shen-Pei Wang, Xichun Li, Huiwu Luo, Huafu Xu

    The task of scene text recognition involves processing information from two modalities: images and text, thereby requiring models to have the ability to extract features from images and model sequences simultaneously. Although linguistic knowledge greatly aids scene text recognition tasks, the extensive use of language models in sequence modeling and model prediction stages in recent years has made model architectures increasingly complex and inefficient. In this paper, we propose LCSTR, a pure convolutional visual model that can complete text recognition without the need for attention mechanisms or language models. This approach applies large kernels to text recognition tasks for the first time, extracting word-level text information through large text-aware blocks, capturing long-range dependencies between characters, and using small text-aware blocks to obtain local features within characters. Experiments show that this model strikes a good trade-off between accuracy and speed, achieving notable results on seven public benchmarks, validating the generalizability and effectiveness of this method. Furthermore, owing to the absence of a language module, this model demonstrates remarkable accuracy even in limited sample scenarios, and the lightweight and low computational overhead features make it suitable for engineering applications.


  • Secure Data Transmission of Smart Home Networks Based on Information Hiding

    by Haiyu Deng, Lina Yang, Xiaocui Dang, Yuan Yan Tang, Patrick Wang

  • Rotation-Invariant Face Detection with Multi-task Progressive Calibration Networks

    by Li-Fang Zhou, Yu Gu, Patrick S. P. Wang, Fa-Yuan Liu, Jie Liu, Tian-Yu Xu