As the workload of radiologists continues to increase, writing radiology reports remains a time-consuming and error-prone task. In recent years, Automated Radiology Report Generation (ARRG) has emerged as a research hotspot aimed at addressing this challenge. This review analyses the main ARRG research streams, including template-based, retrieval-based, encoder-decoder, foundation-model-based, and hybrid approaches. We trace the evolution of ARRG from early template and retrieval paradigms to encoder-decoder and more recent foundation-model-based approaches, while also discussing the growing roles of multimodal, knowledge integration and reinforcement learning strategies, and we compare their respective strengths and limitations. We further summarise the commonly used public, restricted, and private datasets in ARRG research, while distinguishing between datasets that can directly support report generation and auxiliary resources mainly used for pretraining, grounding, or evaluation. In addition, we examine the clinical implications of radiology report format, with particular attention to the trade-offs between free-text and structured reporting and their consequences for model design. We also review mainstream evaluation methods for ARRG, including quantitative metrics (e.g., NLG and CE metrics) and qualitative assessment, and discuss why factual correctness, report organization, and clinical usefulness are not fully captured by surface-level language similarity alone. Finally, we discuss ethical and governance issues that are especially salient for ARRG, such as privacy, bias, hallucination, omission of key abnormalities, negation errors, and responsibility allocation in clinical workflows. We hope that this review will serve as a useful reference for future ARRG research and for the safe translation of these systems into clinical practice.

Single-camera markerless systems have emerged as a robust methodology for human motion capture and rehabilitation applications. Traditional methodologies typically necessitate multiple strategically positioned cameras or special equipment, including sensors to capture patient ambulatory motion, requiring preliminary calibration and synchronization procedures, which may incur significant costs. This paper presents a wireless single-camera markerless framework for rehabilitation applications that leverages advanced deep learning (DL) architectures to estimate and extract three-dimensional skeletal coordinates from monocular camera views of ambulatory patients. The extracted skeletal representation is subsequently transmitted across wireless communication channels. Then, the rendering technique has been applied for displaying virtual movement of the patient for privacy enhancement. Simulations demonstrate the effectiveness of the framework while maintaining motion assessment capabilities, presenting opportunities for deployment in remote healthcare monitoring scenarios.

Most existing hyperspectral image (HSI) classification methods primarily focus on capturing subtle spectral variations by leveraging local spectral-spatial cues derived from patch-level representations. However, limited attention has been given to exploring the global spatial contextual correlations among pixels of HSI. In this study, we propose the Superpixel Segmentation and Classification Mutual Enhancement Network (S2CMEN), a novel framework that integrates global spatial correlations with spectral information through the mutual enhancement of superpixel segmentation and classification. Specifically, a global spatial adaptive module (GSAM) is designed to obtain the direct correlation of the global classes in HSI. It consists of an Adaptive Spectral-Superpixel Network (ASSN) and a Graph Convolutional Network (GCN), forming a synergistic architecture that effectively captures global spatial relationships by adaptively deriving superpixel results from HSIs. Notably, GSAM offers a transferable global spatial representation for HSI tasks, enabling integration with other spectral feature extraction models. Furthermore, we develop a Spatial-Spectral Fusion Module (SSFM) to obtain comprehensive spectral features and fuse them with the extracted global spatial features. Finally, under the constraint of a unit loss, the Mutual Enhancement Strategy (MES) can make the superpixel segmentation loss and the classification loss mutually enhance each other for better performance. We conducted extensive experiments on three public datasets. The proposed S2CMEN achieves overall classification accuracies of 97.38%, 92.33%, and 91.38% on Indian Pines, Pavia University, and Houston, respectively, consistently surpassing existing state-of-the-art methods.

Segmentation is a critical step in analyzing the developing human fetal brain. There have been vast improvements in automatic segmentation methods in the past several years, and the Fetal Brain Tissue Annotation (FeTA) Challenge 2021 helped to establish an excellent standard of fetal brain segmentation. However, FeTA 2021 was a single center study, limiting real-world clinical applicability and acceptance. The multi-center FeTA Challenge 2022 focused on advancing the generalizability of fetal brain segmentation algorithms for magnetic resonance imaging (MRI). In FeTA 2022, the training dataset contained images and corresponding manually annotated multi-class labels from two imaging centers, and the testing data contained images from these two centers as well as two additional unseen centers. The multi-center data included different MR scanners, imaging parameters, and fetal brain super-resolution algorithms applied. 16 teams participated and 17 algorithms were evaluated. Here, the challenge results are presented, focusing on the generalizability of the submissions. Both in- and out-of-domain, the white matter and ventricles were segmented with the highest accuracy (Top Dice scores: 0.89, 0.87 respectively), while the most challenging structure remains the grey matter (Top Dice score: 0.75) due to anatomical complexity. The top 5 average Dices scores ranged from 0.81-0.82, the top 5 average percentile Hausdorff distance values ranged from 2.3-2.5mm, and the top 5 volumetric similarity scores ranged from 0.90-0.92. The FeTA Challenge 2022 was able to successfully evaluate and advance generalizability of multi-class fetal brain tissue segmentation algorithms for MRI and it continues to benchmark new algorithms.

In this paper, we introduce the healthcare system where Semantic Communication (SC) technology is applied to improve the quality of service for healthcare and medical applications. We first show the concepts and possible architecture of SC. Then, we show different types of SC in the healthcare system. Next, some examples of SC-enhanced healthcare applications are discussed. Finally, we give research challenges and future research directions.

This study provides a comparative analysis of deep learning models—UNet, Res-UNet, Attention Res-UNet, and nnUNet—evaluating their performance in brain tumour, polyp, and multi-class heart segmentation tasks. The analysis focuses on precision, accuracy, recall, Dice Similarity Coefficient (DSC), and Intersection over Union (IoU) to assess their clinical applicability. In brain tumour segmentation, Res-UNet and nnUNet significantly outperformed UNet, with Res-UNet leading in DSC and IoU scores, indicating superior accuracy in tumour delineation. Meanwhile, nnUNet excelled in recall and accuracy, which are crucial for reliable tumour detection in clinical diagnosis and planning. In polyp detection, nnUNet was the most effective, achieving the highest metrics across all categories and proving itself as a reliable diagnostic tool in endoscopy. In the complex task of heart segmentation, Res-UNet and Attention Res-UNet were outstanding in delineating the left ventricle, with Res-UNet also leading in right ventricle segmentation. nnUNet was unmatched in myocardium segmentation, achieving top scores in precision, recall, DSC, and IoU. The conclusion notes that although ResUNet occasionally outperforms nnUNet in specific metrics, the differences are quite small. Moreover, nnUNet consistently shows superior overall performance across the experiments. Particularly noted for its high recall and accuracy, which are crucial in clinical settings to minimize misdiagnosis and ensure timely treatment, nnUNet’s robust performance in crucial metrics across all tested categories establishes it as the most effective model for these varied and complex segmentation tasks.

Virtual try-on models have been developed using deep learning techniques to transfer clothing product images onto a candidate. While previous research has primarily focused on enhancing the realism of the garment transfer, such as improving texture quality and preserving details, there is untapped potential to further improve the shopping experience for consumers. The present study outlines the development of an innovative multi-pose virtual try-on model, namely StyleVTON, to potentially enhance consumers’ shopping experiences. Our method synthesises a try-on image while also allowing for changes in pose. To achieve this, StyleVTON first predicts the segmentation of the target pose based on the target garment. Next, the segmentation layout guides the warping process of the target garment. Finally, the pose of the candidate is transferred to the desired posture. Our experiments demonstrate that StyleVTON can generate satisfactory images of candidates wearing the desired clothes in a desired pose, potentially offering a promising solution for enhancing the virtual try-on experience. Our findings reveal that StyleVTON outperforms other comparable methods, particularly in preserving the facial identity of the candidate and geometrically transforming the garments.

We introduce a novel image-based virtual try-on model designed to replace a candidate's garment with a desired target item.The proposed model comprises three modules: segmentation, garment warping, and candidate-clothing fusion.Previous methods have shown limitations in cases involving significant differences between the original and target clothing, as well as substantial overlapping of body parts.Our model addresses these limitations by employing two key strategies.Firstly, it utilises a candidate representation based on an RGB skeleton image to enhance spatial relationships among body parts, resulting in robust segmentation and improved occlusion handling.Secondly, truncated U-Net is employed in both the segmentation and warping modules, enhancing segmentation performance and accelerating the try-on process.The warping module leverages an efficient affine transform for ease of training.Comparative evaluations against state-of-the-art models demonstrate the competitive performance of our proposed model across various scenarios, particularly excelling in handling occlusion cases and significant differences in clothing cases.This research presents a promising solution for image-based virtual try-on, advancing the field by overcoming key limitations and achieving superior performance.

The nnUNet is a fully automated and generalisable framework which automatically configures the full training pipeline for the segmentation task it is applied on, while taking into account dataset properties and hardware constraints. It utilises a basic UNet type architecture which is self-configuring in terms of topology. In this work, we propose to extend the nnUNet by integrating mechanisms from more advanced UNet variations such as the residual, dense, and inception blocks, resulting in three new nnUNet variations, namely the Residual-nnUNet, Dense-nnUNet, and Inception-nnUNet. We have evaluated the segmentation performance on eight datasets consisting of 20 target anatomical structures. Our results demonstrate that altering network architecture may lead to performance gains, but the extent of gains and the optimally chosen nnUNet variation is dataset dependent.

2D based Virtual Try-On (VTON) has been trending towards using human parsing to improve the quality of the try-on image. However, it remains a challenging problem for most existing VTON models to generate realistic images for situations with unpaired candidate-clothing images and body-part occlusions. We have developed a Simplified Virtual Try-On (SVTON) model to rectify the above problem. The SVTON uses refined input data to produce accurate labels and has fewer trainable parameters than existing methods. Also, it is designed with a simplified network architecture for segmentation and an efficient Affine Transform for warping to target clothing. Experiments on benchmark datasets show that the proposed model performs better than the state-of-the-art VTON models for unpaired and occlusion cases, while maintaining the similar overall performance level for normal cases.

Retinal image analysis is increasingly important for diagnosing eye diseases, and blood vessels are one of the most important indicators. This paper presents an automated and unsupervised method for segmenting retinal blood vessels from fundus images by using the level set method, which adopts ChanVese region-based term with a Gaussian mixture term and a distance regularisation term. Also included in the method are the morphological closing operation and matched filtering to preserve the vessels inside the optic disc and remove the noise of the optic disc boundary, and to enhance the blood vessel information. The effectiveness of this method is demonstrated through testing and comparing with the state-of-the-art methods on three public datasets DRIVE, STARE and HRF. The experimental results show that our method offers several advantages over other methods, in particular in dealing with interference from the optic disc, segmenting vessels inside the optic disc and segmenting small vessel branches.

Optical coherence tomography (OCT) is a noninvasive imaging modality that provides in-depth images of the retina. Properties of individual layers on OCT have become important markers for diagnosing and tracking medication of various eye diseases in current ophthalmology. Manual segmentation of OCT scans posed many challenges (errors, inconsistency), which can be addressed by automated segmentation methods. Level set method is one of the most popular methods in the literature used for this purpose. Although level set methods have a fundamental way of handling topological changes, the weak boundaries and noise in addition to inhomogeneity in OCT images make it difficult to segment the layers accurately. Inspired by the concept of region competition, we incorporate prior knowledge of the retinal structure to segment nine (9) layers of the retina. Mainly, we establish a specific region of interest, then use selected components from fuzzy C-Means for initialisation. The clustering in the initialisation stage is also used to guide the evolution through; a Mumford-Shah (MS) selective region competition force and a Hamilton-Jacobi (HJ) balloon force. The forces ensure evolution close to actual retinal boundaries. Finally, the convergence of the method is based on an improved HJ object indication function influenced by the fuzzy membership to prevent leakages at weak boundaries. Experimental results are promising based on 200 OCT images.

Breast cancer is the most prevalent disease to females in the worldwide. Its pathology remains unclear. Genetics factors is the ways to understand the molecular mechanism. This paper proposed a computational approach to explore the interactions of genes2genes related to breast cancer. We first defined the interactions of genes2genes, and described the representation of interactions of genes2genes. Using the experimental dataset, we implemented the proposed approach for extracting the interactions of genes2genes. Moreover, we also represented the interactions of genes2genes in two forms: relationship matrix and network visualization. By manual analysis, we extracted the interactions of top 10 genes2genes is related to breast cancer, which show the approach is promising for studying molecular mechanism related to breast cancer.

The segmentation of retinal layers is vital for tracking progress of medication and diagnosis of various eye diseases. To date many methods for the analysis exist, however the speckle noise and shadows of retinal blood vessel remains a challenge, with negative influence on the performance of segmentation algorithms. Previous attempts have been focused on image preprocessing or developing sophisticated models for segmentation to address this problem, but it still remains an area of active research. In this paper we propose a simple yet efficient and computationally inexpensive method by using fuzzy histogram hyperbolization for enhancement technique, and continuous max-flow for segmentation of four retinal layers (Inner Limiting membrane, Retinal Nerve Fibre Layer, Outer segment and the Retinal Pigment Epithelium). The results show improvement in segmentation performance.

Malignant melanoma is one of the most rapidly increasing cancers globally and it is the most dangerous form of human skin cancer. Dermoscopy is one of the major imaging modalities used in the diagnosis of melanoma. Early detection of melanoma can be helpful and usually curable. Due to the difficulty for dermatologists in the interpretation of dermoscopy images, Computer Aided Diagnosis systems can be very helpful to facilitate the early detection. The automated detection of the lesion borders is one of the most important steps in dermoscopic image analysis. In this paper, we present a fully automated method for melanoma border detection using image processing techniques. The hair and several noises aredetected and removed by applying a bank of directional filters and Image Inpainting method respectively. A hybrid method is developed by combining Particle Swarm Optimization and Markov Random Field methods, in order to delineate the border of the lesion area in the images. The method was tested on a dataset of 200 dermoscopic images, and the experimental results show that our method is superior in terms of the accuracy of drawing the lesion borders compared to alternative methods.

The choroid is an important vascular layer that supplies oxygen and nourishment to the retina. The changes in thickness of the choroid have been hypothesized to relate to a number of retinal diseases in the pathophysiology. In this paper, an automatic method is proposed for segmenting the choroidal layer from macular images by using the level set framework. The three-dimensional nonlinear anisotropic diffusion filter is used to remove all the optical coherence tomography (OCT) imaging artifacts including the speckle noise and to enhance the contrast. The distance regularization and edge constraint terms are embedded into the level set method to avoid the irregular and small regions and keep information about the boundary between the choroid and sclera. Besides, the Markov random field method models the region term into the framework by correlating the single-pixel likelihood function with neighborhood information to compensate for the inhomogeneous texture and avoid the leakage due to the shadows cast by the blood vessels during imaging process. The effectiveness of this method is demonstrated by comparing against other segmentation methods on a dataset with manually labeled ground truth. The results show that our method can successfully and accurately estimate the posterior choroidal boundary.

Abstract Head‐mounted displays with dense pixel arrays used for virtual reality applications require high frame rates and low latency rendering. This forms a challenging use case for any rendering approach. In addition to its ability of generating realistic images, ray tracing offers a number of distinct advantages, but has been held back mainly by its performance. In this paper, we present an approach that significantly improves image generation performance of ray tracing. This is done by combining foveated rendering based on eye tracking with reprojection rendering using previous frames in order to drastically reduce the number of new image samples per frame. To reproject samples a coarse geometry is reconstructed from a G‐Buffer. Possible errors introduced by this reprojection as well as parts that are critical to the perception are scheduled for resampling. Additionally, a coarse color buffer is used to provide an initial image, refined smoothly by more samples were needed. Evaluations and user tests show that our method achieves real‐time frame rates, while visual differences compared to fully rendered images are hardly perceivable. As a result, we can ray trace non‐trivial static scenes for the Oculus DK2 HMD at 1182 × 1464 per eye within the the VSync limits without perceived visual differences.

Retinal image analysis is increasingly prominent as a nonintrusive diagnosis method in modern ophthalmology. In this paper, we present a novel method to segment blood vessels and optic disk in the fundus retinal images. The method could be used to support nonintrusive diagnosis in modern ophthalmology since the morphology of the blood vessel and the optic disk is an important indicator for diseases like diabetic retinopathy, glaucoma, and hypertension. Our method takes as first step the extraction of the retina vascular tree using the graph cut technique. The blood vessel information is then used to estimate the location of the optic disk. The optic disk segmentation is performed using two alternative methods. The Markov random field (MRF) image reconstruction method segments the optic disk by removing vessels from the optic disk region, and the compensation factor method segments the optic disk using the prior local intensity knowledge of the vessels. The proposed method is tested on three public datasets, DIARETDB1, DRIVE, and STARE. The results and comparison with alternative methods show that our method achieved exceptional performance in segmenting the blood vessel and optic disk.

This study addresses the robust set-membership finite-horizon filtering problem for a class of discrete time-varying systems with missing measurement and polytopic uncertainties in the presence of unknown-but-bounded process and measurement noises. A robust set-membership filter is developed and a recursive algorithm is derived for computing the state estimate ellipsoid that is guaranteed to contain the true state. An optimal possible estimate set is computed recursively by solving the semi-definite programming problem. Simulation results are provided to demonstrate the effectiveness of the proposed method.

Image analysis is becoming increasingly prominent as a non intrusive diagnosis in modern ophthalmology. Blood vessel morphology is an important indicator for diseases like diabetes, hypertension and retinopathy. This paper presents an automated and unsupervised method for retinal blood vessels segmentation using the graph cut technique. The graph is constructed using a rough segmentation from a pre-processed image together with spatial pixel connection. The proposed method was tested on two public datasets and compared with other methods. Experimental results show that this method outperforms other unsupervised methods and demonstrate the competitiveness with supervised methods.

Traffic sign recognition is an example of a hard multiclass classification problem. The existing approaches to that problem typically associate with each sign class a real-valued likelihood function and assign such a label to the unknown image that maximizes the value of this function. These template-matching techniques are usually based on arbitrary similarity metrics, such as normalized cross correlation, which do not capture the characteristics of the sign imagery. In this paper, we study the concept of a robust sign similarity measure that can be inferred from the domain-specific data. Two novel machine-learning techniques are proposed as a framework for automatic construction of such a measure from the pairs of images representing either the same or different classes. One is called SimBoost, which is a variation of the AdaBoost algorithm, and the other is based on the fuzzy regression tree framework. Through the experiments with low-quality images, we show that the proposed method admits efficient road sign recognition and outperforms the existing approaches in terms of the classification accuracy.

This paper is concerned with the set-membership filtering (SMF) problem for discrete-time nonlinear systems. We employ the Takagi-Sugeno (T-S) fuzzy model to approximate the nonlinear systems over the true value of state and to overcome the difficulty with the linearization over a state estimate set rather than a state estimate point in the set-membership framework. Based on the T-S fuzzy model, we develop a new nonlinear SMF estimation method by using the fuzzy modeling approach and the S-procedure technique to determine a state estimation ellipsoid that is a set of states compatible with the measurements, the unknown-but-bounded process and measurement noises, and the modeling approximation errors. A recursive algorithm is derived for computing the ellipsoid that guarantees to contain the true state. A smallest possible estimate set is recursively computed by solving the semidefinite programming problem. An illustrative example shows the effectiveness of the proposed method for a class of discrete-time nonlinear systems via fuzzy switch.

In this letter, an error square constrained filtering problem is considered for systems with both non-Gaussian noises and polytopic uncertainty. A novel filter is developed to estimate the systems states based on the current observation and known deterministic input signals. A free parameter is introduced in the filter to handle the uncertain input matrix in the known deterministic input term. In addition, unlike the existing variance constrained filters, which are constructed by the previous observation, the filter is formed from the current observation. A time-varying linear matrix inequality (LMI) approach is used to derive an upper bound of the state estimation error square. The optimal bound is obtained by solving a convex optimization problem via semi-definite programming (SDP) approach. Simulation results are provided to demonstrate the effectiveness of the proposed method.

Real-time road sign recognition has been of great interest for many years. This problem is often addressed in a two-stage procedure involving detection and classification. In this paper a novel approach to sign representation and classification is proposed. In many previous studies focus was put on deriving a set of discriminative features from a large amount of training data using global feature selection techniques e.g. Principal Component Analysis or AdaBoost. In our method we have chosen a simple yet robust image representation built on top of the Colour Distance Transform (CDT). Based on this representation, we introduce a feature selection algorithm which captures a variable-size set of local image regions ensuring maximum dissimilarity between each individual sign and all other signs. Experiments have shown that the discriminative local features extracted from the template sign images enable minimum-distance classification with error rate not exceeding 7%. 1

Principal component analysis (PCA) is a well-established technique in image processing and pattern recognition. Incremental PCA and robust PCA are two interesting problems with numerous potential applications. However, these two issues have only been separately addressed in the previous studies. In this paper, we present a novel algorithm for incremental and robust PCA by seamlessly integrating the two issues together. The proposed algorithm has the advantages of both incremental PCA and robust PCA. Moreover, unlike most M-estimation based robust algorithms, it is computational efficient. Experimental results on dynamic background modelling are provided to show the performance of the algorithm with a comparison to the conventional batch-mode and nonrobust algorithms.

A novel approach to image mosaicking from MPEG video is presented in this paper. The motion vectors in both P- and B-frames are used for global motion estimation. The bi-directional information in B-frames provides multiple routes to warp a frame to its previous anchor frame. A least median of squares based algorithm is adopted for robust motion estimation. In the case of a large proportion of outliers, we detect possible algorithm failure and perform re-estimation along a different route. Based on the motion parameters between consecutive frames, the static background panorama and dynamic foreground panorama are constructed from warped images over a whole video sequence.

We investigate the problem of automated video classification by analysing the low-level audio-visual signal patterns along the time course in a holistic manner. Five popular TV broadcast genre are studied including sports, cartoon, news, commercial and music. A novel statistically based approach is proposed comprising two important ingredients designed for implicit semantic content characterisation and class identities modelling. First, a spatial-temporal audio-visual "concatenated" feature vector is composed, aiming to capture crucial clip-level video structure information inherent in a video genre. Second, the feature vector is further processed using principal component analysis to reduce the spatial-temporal redundancy while exploiting the correlations between feature elements. This gives rise to a compact representation fro effective probabilistic modelling of each video genre. Extensive experiments are conducted assessing various aspects of the approach and their influence on the overall system performance.

We present a comprehensive approach to address three challenging problems in face recognition: modelling faces across multi-views, extracting the nonlinear discriminating features, and recognising moving faces dynamically in image sequences. A multi-view dynamic face model is designed to extract the shape-and-pose-free facial texture patterns. Kernel discriminant analysis, which employs the kernel technique to perform linear discriminant analysis in a high-dimensional feature space, is developed to extract the significant nonlinear features which maximise the between-class variance and minimise the within-class variance. Finally, an identity surface based face recognition is performed dynamically from video input by matching object and model trajectories. q 2003 Elsevier B.V. All rights reserved.

Recognising face with large pose variation is more challenging than that in a fixed view, e.g. frontal-view, due to the severe non-linearity caused by rotation in depth, self-shading and self-occlusion. To address this problem, a multi-view dynamic face model is designed to extract the shape-and-pose-free facial texture patterns from multi-view face images. Kernel Discriminant Analysis is developed to extract the significant non-linear discriminating features which maximise the between-class variance and minimise the within-class variance. By using the kernel technique, this process is equivalent to a Linear Discriminant Analysis in a high-dimensional feature space which can be solved conveniently. The identity surfaces are then constructed from these non-linear discriminating features. Face recognition can be performed dynamically from an image sequence by matching an object trajectory and model trajectories on the identity surfaces.

We present an integrated framework for dynamic face detection and recognition, where head pose is estimated using Support Vector Regression, face detection is performed by Support Vector Classification, and recognition is carried out in a feature space constructed by Linear Discriminant Analysis. Unlike most traditional approaches to matching the patterns from static face images, we model the dynamics of human faces from video sequences in a consistent spatio-temporal context, i.e. recognition is accomplished by matching an object trajectory to a set of identity model trajectories in feature space. The model trajectories are synthesized from only a few views which sparsely cover the view sphere. Compared with the static face matching techniques, this approach is more robust and accurate under a coarse correspondence of face images, and has potential to visual interaction and advanced human behaviour recognition in real-world scenarios. 1 Introduction The issue of face rec...

A support vector machine-based multi-view face detection and recognition framework is described. Face detection is carried out by constructing several detectors, each of them in charge of one specific view. The symmetrical property of face images is employed to simplify the complexity of the modelling. The estimation of head pose, which is achieved by using the support vector regression technique, provides crucial information for choosing the appropriate face detector. This helps to improve the accuracy and reduce the computation in multi-view face detection compared to other methods. For video sequences, further computational reduction can be achieved by using a pose change smoothing strategy. When face detectors find a face in frontal view, a support vector machine-based multi-class classifier is activated for face recognition. All the above issues are integrated under a support vector machine framework. Test results on four video sequences are presented, among them the detection rate is above 95%, recognition accuracy is above 90%, average pose estimation error is around 10/spl deg/, and the full detection and recognition speed is up to 4 frames/second on a Pentium II 300 PC.

An approach to multi-view face detection based on head pose estimation is presented in this paper. Support vector regression is employed to solve the problem of pose estimation. Three methods, the eigenface method the support vector machine (SVM) based method, and a combination of the two methods, are investigated. The eigenface method, which seeks to estimate the overall probability distribution of patterns to be recognised, is fast but less accurate because of the overlap of confidence distributions between face and non-face classes. On the other hand, the SVM method, which tries to model the boundary of two classes to be classified is more accurate but slower as the number of support vectors is normally large. The combined method can achieve an improved performance by speeding up the computation and keeping the accuracy to a preset level. It can be used to automatically detect and track faces in face verification and identification systems.