Juan Carlos Moreno

Fast and globally convex multiphase active contours for brain MRI segmentation

Abstract Multiphase active contour based models are useful in identifying multiple regions with spatial consistency but varying characteristics such as the mean intensities of regions. Segmenting brain magnetic resonance images (MRIs) using a multiphase approach is useful to differentiate white and gray matter tissue for anatomical, functional and disease studies. Multiphase active contour methods are superior to other approaches due to their topological flexibility, accurate boundaries, robustness to image variations and adaptive energy functionals. Globally convex methods are furthermore initialization independent. We extend the relaxed globally convex Chan and Vese two-phase piecewise constant energy minimization formulation of Chan et al. (2006) [1] to the multiphase domain and prove the existence of a global minimizer in a specific space which is one of the novel contributions of the paper. An efficient dual minimization implementation of our binary partitioning function model accurately describes disjoint regions using stable segmentations by avoiding local minima solutions. Experimental results indicate that the proposed approach provides consistently better accuracy than other related multiphase active contour algorithms using four different error metrics (Dice, Rand Index, Global Consistency Error and Variation of Information) even under severe noise, intensity inhomogeneities, and partial volume effects in MRI imagery.Multiphase active contour based models are useful in identifying multiple regions with different characteristics such as the mean values of regions. This is relevant in brain magnetic resonance images (MRIs), allowing the differentiation of white matter against gray matter. We consider a well defined globally convex formulation of Vese and Chan multiphase active contour model for segmenting brain MRI images. A well-established theory and an efficient dual minimization scheme are thoroughly described which guarantees optimal solutions and provides stable segmentations. Moreover, under the dual minimization implementation our model perfectly describes disjoint regions by avoiding local minima solutions. Experimental results indicate that the proposed approach provides better accuracy than other related multiphase active contour algorithms even under severe noise, intensity inhomogeneities, and partial volume effects.

Acquiring high-resolution face images in outdoor environments: A master-slave calibration algorithm

Facial recognition at-a-distance in surveillance scenarios remains an open problem, particularly due to the small number of pixels representing the facial region. The use of pan-tilt-zoom (PTZ) cameras has been advocated to solve this problem, however, the existing approaches either rely on rough approximations or additional constraints to estimate the mapping between image coordinates and pan-tilt parameters. In this paper, we aim at extending PTZ-assisted facial recognition to surveillance scenarios by proposing a master-slave calibration algorithm capable of accurately estimating pan-tilt parameters without depending on additional constraints. Our approach exploits geometric cues to automatically estimate subjects height and thus determine their 3D position. Experimental results show that the presented algorithm is able to acquire high-resolution face images at a distance ranging from 5 to 40 meters with high success rate. Additionally, we certify the applicability of the aforementioned algorithm to biometric recognition through a face recognition test, comprising 20 probe subjects and 13,020 gallery subjects.

A segmentation model and application to endoscopic images

In this paper a variational segmentation model is proposed. It is a generalization of the Chan and Vese model, for the scalar and vector-valued cases. It incorporates extra terms, depending on the image gradient, and aims at approximating the smoothed image gradient norm, inside and outside the segmentation curve, by mean constant values. As a result, a flexible model is obtained. It segments, more accurately, any object displaying many oscillations in its interior. In effect, an external contour of the object, as a whole, is achieved, together with internal contours, inside the object. For determining the approximate solution a Levenberg-Marquardt Newton-type optimization method is applied to the finite element discretization of the model. Experiments on in vivo medical endoscopic images (displaying aberrant colonic crypt foci) illustrate the efficacy of this model.

Robust periocular recognition by fusing local to holistic sparse representations

Sparse representations have been advocated as a relevant advance in biometrics research. In this paper we propose a new algorithm for fusion at the data level of sparse representations, each one obtained from image patches. The main novelties are two-fold: 1) a dictionary fusion scheme is formalised, using the l1--- minimization with the gradient projection method; 2) the proposed representation and classification method does not require the non-overlapping condition of image patches from where individual dictionaries are obtained. In the experiments, we focused in the recognition of periocular images and obtained independent dictionaries for the eye, eyebrow and skin regions, that were subsequently fused. Results obtained in the publicly available UBIRIS.v2 data set show consistent improvements in the recognition effectiveness when compared to state-of-the-art related representation and classification techniques.

Feature preserving anisotropic diffusion for image restoration

Anisotropic diffusion based schemes are widely used in image smoothing and noise removal. Typically, the partial differential equation (PDE) used is based on computing image gradients or isotropically smoothed version of the gradient image. To improve the denoising capability of such nonlinear anisotropic diffusion schemes, we introduce a multi-direction based discretization along with a selection strategy for choosing the best direction of possible edge pixels. This strategy avoids the directionality based bias which can over-smooth features that are not aligned with the coordinate axis. The proposed hybrid discretization scheme helps in preserving multi-scale features present in the images via selective smoothing of the PDE. Experimental results indicate such an adaptive modification provides improved restoration results on noisy images.

On convergent finite difference schemes for variational–PDE-based image processing

We study an adaptive anisotropic Huber functional-based image restoration scheme. Using a combination of L2–L1 regularization functions, an adaptive Huber functional-based energy minimization model provides denoising with edge preservation in noisy digital images. We study a convergent finite difference scheme based on continuous piecewise linear functions and use a variable splitting scheme, namely the Split Bregman (In: Goldstein and Osher, SIAM J Imaging Sci 2(2):323–343, 2009) algorithm, to obtain the discrete minimizer. Experimental results are given in image denoising and comparison with additive operator splitting, dual fixed point, and projected gradient schemes illustrates that the best convergence rates are obtained for our algorithm.

Joint Head Pose/Soft Label Estimation for Human Recognition In-The-Wild

Soft biometrics have been emerging to complement other traits and are particularly useful for poor quality data. In this paper, we propose an efficient algorithm to estimate human head poses and to infer soft biometric labels based on the 3D morphology of the human head. Starting by considering a set of pose hypotheses, we use a learning set of head shapes synthesized from anthropometric surveys to derive a set of 3D head centroids that constitutes a metric space. Next, representing queries by sets of 2D head landmarks, we use projective geometry techniques to rank efficiently the joint 3D head centroids/pose hypotheses according to their likelihood of matching each query. The rationale is that the most likely hypotheses are sufficiently close to the query, so a good solution can be found by convex energy minimization techniques. Once a solution has been found, the 3D head centroid and the query are assumed to have similar morphology, yielding the soft label. Our experiments point toward the usefulness of the proposed solution, which can improve the effectiveness of face recognizers and can also be used as a privacy-preserving solution for biometric recognition in public environments.

A Calibration Algorithm for Multi-camera Visual Surveillance Systems Based on Single-View Metrology

The growing concerns about persons security and the increasing popularity of pan-tilt-zoom (PTZ) cameras, have been raising the interest on automated master-slave surveillance systems. Such systems are typically composed by (1) a fixed wide-angle camera that covers a large area, detects and tracks moving objects in the scene; and (2) a PTZ camera, that provides a close-up view of an object of interest. Previously published approaches attempted to establish 2D correspondences between the video streams of both cameras, which is a ill-posed formulation due to the absence of depth information. On the other side, 3D-based approaches are more accurate but require more than one fixed camera to estimate depth information. In this paper, we describe a novel method for easy and precise calibration of a master-slave surveillance system, composed by a single fixed wide-angle camera. Our method exploits single view metrology to infer 3D data of the tracked humans and to self-perform the transformation between camera views. Experimental results in both simulated and realistic scenes point for the effectiveness of the proposed model in comparison with the state-of-the-art.

A Master-Slave Calibration Algorithm with Fish-Eye Correction

Surveillance systems capable of autonomously monitoring vast areas are an emerging trend, particularly when wide-angle cameras are combined with pan-tilt-zoom (PTZ) cameras in a master-slave configuration. The use of fish-eye lenses allows the master camera to maximize the coverage area while the PTZ acts as a foveal sensor, providing high-resolution images of regions of interest. Despite the advantages of this architecture, the mapping between image coordinates and pan-tilt values is the major bottleneck in such systems, since it depends on depth information and fish-eye effect correction. In this paper, we address these problems by exploiting geometric cues to perform height estimation. This information is used both for inferring 3D information from a single static camera deployed on an arbitrary position and for determining lens parameters to remove fish-eye distortion. When compared with the previous approaches, our method has the following advantages: (1) fish-eye distortion is corrected without relying on calibration patterns; (2) 3D information is inferred from a single static camera disposed on an arbitrary location of the scene.

Robust Periocular Recognition by Fusing Sparse Representations of Color and Geometry Information

In this paper, we propose a re-weighted elastic net (REN) model for biometric recognition. The new model is applied to data separated into geometric and color spatial components. The geometric information is extracted using a fast cartoon - texture decomposition model based on a dual formulation of the total variation norm allowing us to carry information about the overall geometry of images. Color components are defined using linear and nonlinear color spaces, namely the red-green-blue (RGB), chromaticity-brightness (CB) and hue-saturation-value (HSV). Next, according to a Bayesian fusion-scheme, sparse representations for classification purposes are obtained. The scheme is numerically solved using a gradient projection (GP) algorithm. In the empirical validation of the proposed model, we have chosen the periocular region, which is an emerging trait known for its robustness against low quality data. Our results were obtained in the publicly available FRGC and UBIRIS.v2 data sets and show consistent improvements in recognition effectiveness when compared to related state-of-the-art techniques.