Abstract not found

Abstract. Many approaches have been proposed to locate faces in an image. There are, however, two problems in previous facial shape models using feature points. First, the dimension of the solution space is too big since a large number of key points are needed to model a face. Second, the local features associated with the key points are assumed to be independent. Therefore, previous approaches require good initialization (which is often done manually), and may generate inaccurate localization. To automatically locate faces, we propose a novel hierarchical shape model (HSM) or multi-resolution shape models corresponding to a Gaussian pyramid of the face image. The coarsest shape model can be quickly located in the lowest resolution image. The located coarse model is then used to guide the search for a finer face model in the higher resolution image. Moreover, we devise a Global and Local (GL) distribution to learn the likelihood of the joint distribution of facial features. A novel hierarchical data-driven Markov chain Monte Carlo (HDDMCMC) approach is proposed to achieve the global optimum of face localization. Experimental results demonstrate that our algorithm produces accurate localization results quickly, bypassing the need for good initialization. 1

Cloth modeling and recognition is an important and challenging problem in both vision and graphics tasks, such as dressed human recognition and tracking, human sketch and portrait. In this paper, we present a context sensitive grammar in an And-Or graph representation which will produce a large set of composite graphical templates to account for the wide variabilities of cloth configurations, such as T-shirts, jackets, etc. In a supervised learning phase, we ask an artist to draw sketches on a set of dressed people, and we decompose the sketches into categories of cloth and body components: collars, shoulders, cuff, hands, pants, shoes etc. Each component has a number of distinct subtemplates (sub-graphs). These sub-templates serve as leafnodes in a big And-Or graph where an And-node represents a decomposition of the graph into sub-configurations with Markov relations for context and constraints (soft or hard), and an Or-node is a switch for choosing one out of a set of alternative And-nodes (sub-configurations) – similar to a node in stochastic context free grammar (SCFG). This representation integrates the SCFG for structural variability and the Markov (graphical) model for context. An algorithm which integrates the bottom-up proposals and the topdown information is proposed to infer the composite cloth template from the image. 1.

Abstract not found

This paper addresses the 3D tracking of pose and animation of the human face in monocular image sequences using active appearance models. The major problem of the classical appearance-based adaptation is the high computational time resulting from the inclusion of a synthesis step in the iterative optimization. Whenever the dimension of the face subspace is large, a real-time performance cannot be achieved. In this paper, we aim at designing a fast and stable active appearance model search for 3D face tracking. The main contribution is a search algorithm whose CPU-time is not dependent on the dimension of the face subspace. Using this algorithm, we show that both the CPU-time and the likelihood of a non accurate tracking are reduced. Experiments evaluating the effectiveness of the proposed algorithm are reported, as well as method comparison and tracking examples.

In this paper we propose a novel inhomogeneous Gibbs model by the minimax entropy principle, and apply it to face modeling. The maximum entropy principle generalizes the statistical properties of the observed samples and results in the Gibbs distribution, while the minimum entropy principle makes the learnt distribution close to the observed one. To capture the fine details of a face, an inhomogeneous Gibbs model is derived to learn the local statistics of facial feature points. To alleviate the high dimensionality problem of face models, we propose to learn the distribution in a subspace reduced by principal component analysis or PCA. We demonstrate that our model effectively captures important and subtle non-Gaussian face patterns and efficiently generates good face models. 1.

We present a system for finding and tracking a face and extract global and local animation parameters from a video sequence. The system uses an initial colour processing step for finding a rough estimate of the position, size, and inplane rotation of the face, followed by a refinement step driven by an Active Model. The latter step refines the previous estimate, and also extracts local animation parameters. The system is able to track the face and some facial features in near real-time, and can compress the result to a bitstream compliant to MPEG-4 Face & Body Animation. 1.

Appearance Models (AM) are commonly used to model appearance and shape variation of objects in images. In particular, they have proven useful to detection, tracking, and synthesis of people’s faces from video. While AM have numerous advantages relative to alternative approaches, they have at least two important drawbacks. First, they are especially prone to local minima in fitting; this problem becomes increasingly problematic as the number of parameters to estimate grows. Second, often few if any of the local minima correspond to the correct location of the model error. To address these problems, we propose Filtered Component Analysis (FCA), an extension of traditional Principal Component Analysis (PCA). FCA learns an optimal set of filters with which to build a multi-band representation of the object. FCA representations were found to be more robust than either grayscale or Gabor filters to problems of local minima. The effectiveness and robustness of the proposed algorithm is demonstrated in both synthetic and real data.

In this paper, we present a set of extensions to the deformable template model: Active Appearance Model (AAM) proposed by Cootes et al. AAMs distinguish themselves by learning a priori knowledge through observation of shape and texture variation in a training set. This is used to obtain a compact object class description, which can be employed to rapidly search images for new object instances. The proposed extensions concern enhanced shape representation, handling of homogeneous and heterogeneous textures, refinement optimization using Simulated Annealing and robust statistics. Finally, an initialization scheme is designed thus making the usage of AAMs fully automated. Using these extensions it is demonstrated that AAMs can segment bone structures in radiographs, pork chops in perspective images and the left ventricle in cardiovascular magnetic resonance images in a robust, fast and accurate manner. Subpixel landmark accuracy was obtained in two of the three cases.

E#cient content-based image retrieval (CBIR) of biomedical images is a challenging problem. Feature representation algorithms used in indexing medical images on the pathology of interest have to address conflicting goals of reducing feature dimensionality while retaining important and often subtle biomedical features. At the Lister Hill National Center for Biomedical Communications, an intramural R&D division of the U.S. National Library of Medicine, we are developing CBIR prototype for digitized images of a collection of 17,000 cervical and lumbar spine X-rays taken as a part of the second National Health and Nutrition Examination Survey (NHANES II). The vertebra shape e#ectively describes various pathologies identified by medical experts as being consistently and reliably found in the image collection. A suitable shape algorithm must represent shapes in low dimension, be invariant to rotation, translation, and scale transforms, and retain relevant pathology. Additionally, supported similarity algorithms must be useful in retrieving images that are relevant to the queries posed by the intended target community, viz. medical researchers, physicians, etc. This paper describes an evaluation of two popular shape similarity methods from the literature on a set of 250 vertebra boundary shapes. The polygon approximation method achieved a performance score of 55.94% and bettered the Fourier descriptor algorithm which had a performance score of 46.96%.