Multi-model fitting is fundamental for robust geometric estimation in computer vision. However, recent deep learning methods enable parallel model detection but rely on simple architectures that inadequately model spatial relationships. Moreover, current methods typically generate hypotheses only through minimal solvers on randomly sampled points, thus failing to explore the full diversity of the solution space. To address these limitations, we propose a novel Jacobian-based Gaussian uncertainty modeling framework, which analytically propagates covariance through geometric transformations and enables efficient expansion of the hypothesis space with strong theoretical guarantees. We further introduce a Gaussian Hypothesis Generation Network (GHG-Net) to learn global parameter distributions, enabling the generation of diverse and geometrically valid hypotheses. Additionally, our network captures spatial relationships among observations by employing a dynamic graph neural network with a multi-head attention mechanism. This yields more accurate sample and inlier weights, significantly improving the quality of hypothesis generation. Extensive experiments on three representative geometric estimation tasks (i.e. vanishing point detection, fundamental matrix estimation, and homography estimation) demonstrate that our method achieves new state-of-the-art accuracy and stability, while maintaining high computational efficiency.

Semantic context is an important and useful cue for scene parsing in complicated natural images with a substantial amount of variations in objects and the environment. This paper proposes Spatially Constrained Location Prior (SCLP) for effective modelling of global and local semantic context in the scene in terms of inter-class spatial relationships. Unlike existing studies focusing on either relative or absolute location prior of objects, the SCLP effectively incorporates both relative and absolute location priors by calculating object co-occurrence frequencies in spatially constrained image blocks. The SCLP is general and can be used in conjunction with various visual feature-based prediction models, such as Artificial Neural Networks and Support Vector Machine (SVM), to enforce spatial contextual constraints on class labels. Using SVM classifiers and a linear regression model, we demonstrate that the incorporation of SCLP achieves superior performance compared to the state-of-the-art methods on the Stanford background and SIFT Flow datasets.

Roadside vegetation classification has recently attracted increasing attention, due to its significance in applications such as vegetation growth management and fire hazard identification. Existing studies primarily focus on learning visible feature based classifiers or invisible feature based thresholds, which often suffer from a generalization problem to new data. This paper proposes an approach that aggregates pixel-level supervised classification and cluster-level texton occurrence within a voting strategy over superpixels for vegetation classification, which takes into account both generic features in the training data and local characteristics in the testing data. Class-specific artificial neural networks are trained to predict class probabilities for all pixels, while a texton based adaptive K-means clustering process is introduced to group pixels into clusters and obtain texton occurrence. The pixel-level class probabilities and cluster-level texton occurrence are further integrated in superpixel-level voting to assign each superpixel to a class category. The proposed approach outperforms previous approaches on a roadside image dataset collected by the Department of Transport and Main Roads, Queensland, Australia, and achieves state-of-the-art performance using low-resolution images from the Croatia roadside grass dataset.

Classification of viewer interest using facial expression and heart rate facilitates automatic identification of interest evoking video segments. Sports video is suitable for testing the effectiveness of such a system as it has structured segments with distinguishable highlight events. Previous work has not investigated the differences in viewer interest characteristics from one sports type to another, which is crucial for appropriate classification methodology. Thus, it is still unclear whether a universal classification model can be used for analyzing viewer interest for all types of sports. This paper addresses this gap by demonstrating a significant difference (p < 0.05) in the distributions of viewer interest data in soccer compared to tennis. Based on this finding, this paper proposes an adoption of Gaussian mixture models (GMM) to integrate sports-specific and sports-independent approaches for identifying video segments, which would be of potential interest to individual viewers. The approaches achieve 52% to 64% accuracy, demonstrating that sports-specific approach gets better performance.

Viewer interest, evoked by video content, can potentially identify the highlights of the video. This paper explores the use of facial expressions (FE) and heart rate (HR) of viewers captured using camera and non-strapped sensor for identifying interesting video segments. The data from ten subjects with three videos showed that these signals are viewer dependent and not synchronized with the video contents. To address this issue, new algorithms are proposed to effectively combine FE and HR signals for identifying the time when viewer interest is potentially high. The results show that, compared with subjective annotation and match report highlights, 'non-neutral' FE and 'relatively higher and faster' HR is able to capture 60%-80% of goal, foul, and shot-on-goal soccer video events. FE is found to be more indicative than HR of viewer interest, but the fusion of these two modalities outperforms each of them.

The proliferation of news reports published in online websites and news information sharing among social media users necessitates effective techniques for analysing the image, text and video data related to news topics. This paper presents the first study to classify affective facial images on emerging news topics. The proposed system dynamically monitors and selects the current hot (of great interest) news topics with strong affective interestingness using textual keywords in news articles and social media discussions. Images from the selected hot topics are extracted and classified into three categorized emotions, positive, neutral and negative, based on facial expressions of subjects in the images. Performance evaluations on two facial image datasets collected from realworld resources demonstrate the applicability and effectiveness of the proposed system in affective classification of facial images in news reports. Facial expression shows high consistency with the affective textual content in news reports for positive emotion, while only low correlation has been observed for neutral and negative. The system can be directly used for applications, such as assisting editors in choosing photos with a proper affective semantic for a certain topic during news report preparation.

To recognize faces in video, face appearances have been widely modeled as piece-wise local linear models which linearly approximate the smooth yet non-linear low dimensional face appearance manifolds. The choice of representations of the local models is crucial. Most of the existing methods learn each local model individually meaning that they only anticipate variations within each class. In this work, we propose to represent local models as Gaussian distributions which are learned simultaneously using the heteroscedastic probabilistic linear discriminant analysis (PLDA). Each gallery video is therefore represented as a collection of such distributions. With the PLDA, not only the within-class variations are estimated during the training, the separability between classes is also maximized leading to an improved discrimination. The heteroscedastic PLDA itself is adapted from the standard PLDA to approximate face appearance manifolds more accurately. Instead of assuming a single global within-class covariance, the heteroscedastic PLDA learns different within-class covariances specific to each local model. In the recognition phase, a probe video is matched against gallery samples through the fusion of point-to-model distances. Experiments on the Honda and MoBo datasets have shown the merit of the proposed method which achieves better performance than the state-of-the-art technique.

Feature extraction and selection are critical processes in developing facial expression recognition (FER) systems. While many algorithms have been proposed for these processes, direct comparison between texture, geometry and their fusion, as well as between multiple selection algorithms has not been found for spontaneous FER. This paper addresses this issue by proposing a unified framework for a comparative study on the widely used texture (LBP, Gabor and SIFT) and geometric (FAP) features, using Adaboost, mRMR and SVM feature selection algorithms. Our experiments on the Feedtum and NVIE databases demonstrate the benefits of fusing geometric and texture features, where SIFT+FAP shows the best performance, while mRMR outperforms Adaboost and SVM. In terms of computational time, LBP and Gabor perform better than SIFT. The optimal combination of SIFT+FAP+mRMR also exhibits a state-of-the-art performance.

Occlusion is a big challenge for facial expression recognition (FER) in real-world situations. Previous FER efforts to address occlusion suffer from loss of appearance features and are largely limited to a few occlusion types and single testing strategy. This paper presents a robust approach for FER in occluded images and addresses these issues. A set of Gabor based templates is extracted from images in the gallery using a Monte Carlo algorithm. These templates are converted into distance features using template matching. The resulting feature vectors are robust to occlusion. Occluded eyes and mouth regions and randomly places occlusion patches are used for testing. Two testing strategies analyze the effects of these occlusions on the overall recognition performance as well as each facial expression. Experimental results on the Cohn-Kanade database confirm the high robustness of our approach and provide useful insights about the effects of occlusion on FER. Performance is also compared with previous approaches.

Facial expression is an important channel for human communication and can be applied in many real applications. One critical step for facial expression recognition (FER) is to accurately extract emotional features. Current approaches on FER in static images have not fully considered and utilized the features of facial element and muscle movements, which represent static and dynamic, as well as geometric and appearance characteristics of facial expressions. This paper proposes an approach to solve this limitation using "salient” distance features, which are obtained by extracting patch-based 3D Gabor features, selecting the "salient” patches, and performing patch matching operations. The experimental results demonstrate high correct recognition rate (CRR), significant performance improvements due to the consideration of facial element and muscle movements, promising results under face registration errors, and fast processing time. Comparison with the state-of-the-art performance confirms that the proposed approach achieves the highest CRR on the JAFFE database and is among the top performers on the Cohn-Kanade (CK) database.

The design of an A/D converter for the next generation of processes requiring lower supply voltages is explored through the design and implementation of a 12 bit converter in a 1.2 /spl mu/m digital CMOS process using a 2 V analog power supply. The resulting chip is 22.4 mm/sup 2/ and is presently being tested. It is expected to achieve a 1 MHz sampling rate with an analog power dissipation of 60 mW.>