Locally optimized non-local means denoising for low-dose X-ray backscatter imagery.

While recent years have seen considerable progress in image denoising, the leading techniques have been developed for digital photographs or other images that can have very different characteristics than those encountered in X-ray applications. In particular here we examine X-ray backscatter (XBS) images collected by airport security systems, where images are piecewise smooth and edge information is typically more correlated with objects while texture is dominated by statistical noise in the detected signal. In this paper, we show how multiple estimates for a denoised XBS image can be combined using a variational approach, giving a solution that enhances edge contrast by trading off gradient penalties against data fidelity terms. We demonstrate the approach by combining several estimates made using the non-local means (NLM) algorithm, a widely used patch-based denoising method. The resulting improvements hold the potential for improving automated analysis of low-SNR X-ray imagery and can be applied in other applications where edge information is of interest.

The piecewise smooth Mumford-Shah functional on an arbitrary graph.

The Mumford-Shah functional has had a major impact on a variety of image analysis problems, including image segmentation and filtering, and, despite being introduced over two decades ago, it is still in widespread use. Present day optimization of the Mumford-Shah functional is predominated by active contour methods. Until recently, these formulations necessitated optimization of the contour by evolving via gradient descent, which is known for its overdependence on initialization and the tendency to produce undesirable local minima. In order to reduce these problems, we reformulate the corresponding Mumford-Shah functional on an arbitrary graph and apply the techniques of combinatorial optimization to produce a fast, low-energy solution. In contrast to traditional optimization methods, use of these combinatorial techniques necessitates consideration of the reconstructed image outside of its usual boundary, requiring additionally the inclusion of regularization for generating these values. The energy of the solution provided by this graph formulation is compared with the energy of the solution computed via traditional gradient descent-based narrow-band level set methods. This comparison demonstrates that our graph formulation and optimization produces lower energy solutions than the traditional gradient descent based contour evolution methods in significantly less time. Finally, we demonstrate the usefulness of the graph formulation to apply the Mumford-Shah functional to new applications such as point clustering and filtering of nonuniformly sampled images.

Organ segmentation with level sets using local shape and appearance priors.

Organ segmentation is a challenging problem on which recent progress has been made by incorporation of local image statistics that model the heterogeneity of structures outside of an organ of interest. However, most of these methods rely on landmark based segmentation, which has certain drawbacks. We propose to perform organ segmentation with a novel level set algorithm that incorporates local statistics via a highly efficient point tracking mechanism. Specifically, we compile statistics on these tracked points to allow for a local intensity profile outside of the contour and to allow for a local surface area penalty, which allows us to capture fine detail where it is expected. The local intensity and curvature models are learned through landmarks automatically embedded on the surface of the training shapes. We use Parzen windows to model the internal organ intensities as one distribution since this is sufficient for most organs. In addition, since the method is based on level sets, we are able to naturally take advantage of recent work on global shape regularization. We show state-of-the-art results on the challenging problems of liver and kidney segmentation.

Localized priors for the precise segmentation of individual vertebras from CT volume data.

We present algorithms for the automatic and precise segmentation of individual vertebras in CT Volume data. When a local surface evolution method such as the level set is applied to such a complex structure, global shape priors will not be sufficient to avoid the leakage and local minima problems, particularly if precise object boundary is desired. We propose a prior knowledge base that contains localized priors--a group of high-level features whose detection will augment the surface model and be the key to success. Base on this a set of context blockers are applied to prevent the leakages. Carefully designed initial surface when registered with the data helps avoid the local minimum problem. The results of segmentation well approximate the human delineated object boundaries. We also present the validation result of the segmentation of 150 vertebras.

Computerized measurement of facial expression of emotions in schizophrenia.

Deficits in the ability to express emotions characterize several neuropsychiatric disorders and are a hallmark of schizophrenia, and there is need for a method of quantifying expression, which is currently done by clinical ratings. This paper presents the development and validation of a computational framework for quantifying emotional expression differences between patients with schizophrenia and healthy controls. Each face is modeled as a combination of elastic regions, and expression changes are modeled as a deformation between a neutral face and an expressive face. Functions of these deformations, known as the regional volumetric difference (RVD) functions, form distinctive quantitative profiles of expressions. Employing pattern classification techniques, we have designed expression classifiers for the four universal emotions of happiness, sadness, anger and fear by training on RVD functions of expression changes. The classifiers were cross-validated and then applied to facial expression images of patients with schizophrenia and healthy controls. The classification score for each image reflects the extent to which the expressed emotion matches the intended emotion. Group-wise statistical analysis revealed this score to be significantly different between healthy controls and patients, especially in the case of anger. This score correlated with clinical severity of flat affect. These results encourage the use of such deformation based expression quantification measures for research in clinical applications that require the automated measurement of facial affect.