Xosé M. Pardo

Saliency from hierarchical adaptation through decorrelation and variance normalization

This paper presents a novel approach to visual saliency that relies on a contextually adapted representation produced through adaptive whitening of color and scale features. Unlike previous models, the proposal is grounded on the specific adaptation of the basis of low level features to the statistical structure of the image. Adaptation is achieved through decorrelation and contrast normalization in several steps in a hierarchical approach, in compliance with coarse features described in biological visual systems. Saliency is simply computed as the square of the vector norm in the resulting representation. The performance of the model is compared with several state-of-the-art approaches, in predicting human fixations using three different eye-tracking datasets. Referring this measure to the performance of human priority maps, the model proves to be the only one able to keep the same behavior through different datasets, showing free of biases. Moreover, it is able to predict a wide set of relevant psychophysical observations, to our knowledge, not reproduced together by any other model before.

On the relationship between optical variability, visual saliency, and eye fixations: A computational approach

A hierarchical definition of optical variability is proposed that links physical magnitudes to visual saliency and yields a more reductionist interpretation than previous approaches. This definition is shown to be grounded on the classical efficient coding hypothesis. Moreover, we propose that a major goal of contextual adaptation mechanisms is to ensure the invariance of the behavior that the contribution of an image point to optical variability elicits in the visual system. This hypothesis and the necessary assumptions are tested through the comparison with human fixations and state-of-the-art approaches to saliency in three open access eye-tracking datasets, including one devoted to images with faces, as well as in a novel experiment using hyperspectral representations of surface reflectance. The results on faces yield a significant reduction of the potential strength of semantic influences compared to previous works. The results on hyperspectral images support the assumptions to estimate optical variability. As well, the proposed approach explains quantitative results related to a visual illusion observed for images of corners, which does not involve eye movements.

A snake for CT image segmentation integrating region and edge information

Abstract The 3D representation and solid modeling of knee bone structures taken from computed tomography (CT) scans are necessary processes in many medical applications. The construction of the 3D model is generally carried out by stacking the contours obtained from a 2D segmentation of each CT slice, so the quality of the 3D model strongly depends on the precision of this segmentation process. In this work we present a deformable contour method for the problem of automatically delineating the external bone (tibia and fibula) contours from a set of CT scan images. We have introduced a new region potential term and an edge focusing strategy that diminish the problems that the classical snake method presents when it is applied to the segmentation of CT images. We introduce knowledge about the location of the object of interest and knowledge about the behavior of edges in scale space, in order to enhance edge information. We also introduce a region information aimed at complementing edge information. The novelty in that is that the new region potential does not rely on prior knowledge about image statistics; the desired features are derived from the segmentation in the previous slice of the 3D sequence. Finally, we show examples of 3D reconstruction demonstrating the validity of our model. The performance of our method was visually and quantitatively validated by experts.

Decorrelation and Distinctiveness Provide with Human-Like Saliency

In this work, we show the capability of a new model of saliency, of reproducing remarkable psychophysical results. The model presents low computational complexity compared to other models of the state of the art. It is based in biologically plausible mechanisms: the decorrelation and the distinctiveness of local responses. Decorrelation of scales is obtained from principal component analysis of multiscale low level features. Distinctiveness is measured through the Hotelling’s T2 statistic. The model is conceived to be used in a machine vision system, in which attention would contribute to enhance performance together with other visual functions. Experiments demonstrate the consistency with a wide variety of psychophysical phenomena, that are referenced in the visual attention modeling literature, with results that outperform other state of the art models.

Cellular neural networks and active contours: a tool for image segmentation

Abstract In this paper Cellular Neural Networks (CNN) are applied to image segmentation based on active contour techniques. The approach is based on deformable contours which evolve pixel by pixel from their initial shapes and locations until delimiting the objects of interest. The contour shift is guided by external information from the image under consideration which attracts them towards the target characteristics (intensity extremes, edges, etc.) and by internal forces which try to maintain the smoothness of the contour curve. This CNN-based proposal combines the characteristics from implicit and parametric models. As a consequence a high flexibility and control for the evolution dynamics of the snakes are provided, allowing the solution of complex tasks as is the case of the topologic transformations. In addition the proposal is suitable for its implementation as an integrated circuit allowing to take advantages of the massively parallel processing in CNN to reduce processing time.

Feature analysis for human recognition and discrimination: Application to a person-following behaviour in a mobile robot

One of the most important abilities that personal robots need when interacting with humans is the ability to discriminate amongst them. In this paper, we carry out an in-depth study of the possibilities of a colour camera placed on top of a robot to discriminate between humans, and thus get a reliable person-following behaviour on the robot. In particular we have reviewed and analysed the possibility of using the most popular colour and texture features used in object and texture recognition, to identify and model the target (person being followed). Nevertheless, the real-time restrictions make necessary the selection of a reduced subset of these features to reduce the computational burden. This subset of features was selected after carrying out a redundancy analysis, and considering how these features perform when discriminating amongst similar human torsos. Finally, we also describe several scoring functions able to dynamically adjust the relevance of each feature considering the particular conditions of the environment where the robot moves, together with the characteristics of the clothes worn by the persons that are in the scene. The results of this in-depth study have been implemented in a novel and adaptive system (described in this paper), which is able to discriminate between humans to get reliable person-following behaviours in a mobile robot. The performance of our proposal is clearly shown through a set of experimental results obtained with a real robot working in real and difficult scenarios.

Biomedical active segmentation guided by edge saliency

Abstract Deformable models are very popular approaches in biomedical image segmentation. Classical snake models are edge-oriented and work well if the target objects have distinct gradient values. This is not always true in biomedical imagery, which makes the model very dependent on initial conditions. In this work we propose an edge-based potential aimed at the elimination of local minima due to undesired edges. The new approach integrates knowledge about the features of the desired boundaries apart from gradient strength and uses a new method to eliminate local minima, which makes the segmentation less sensitive to initial contours.

Discriminant snakes for 3D reconstruction of anatomical organs.

In this work a new statistic deformable model for 3D segmentation of anatomical organs in medical images is proposed. A statistic discriminant snake performs a supervised learning of the object boundary in an image slice to segment the next slice of the image sequence. Each part of the object boundary is projected in a feature space generated by a bank of Gaussian filters. Then, clusters corresponding to different boundary pieces are constructed by means of linear discriminant analysis. Finally, a parametric classifier is generated from each contour in the image slice and embodied into the snake energy-minimization process to guide the snake deformation in the next image slice. The discriminant snake selects and classifies image features by the parametric classifier and deforms to minimize the dissimilarity between the learned and found image features. The new approach is of particular interest for segmenting 3D images with anisotropic spatial resolution, and for tracking temporal image sequences. In particular, several anatomical organs from different imaging modalities are segmented and the results compared to expert tracings.

Two-view line matching algorithm based on context and appearance in low-textured images

A novel approach for line detection and matching is proposed, aimed at achieving good performance with low-textured scenes, under uncontrolled illumination conditions. Line detection is performed by means of phase-based edge detector over Gaussian scale-space, followed by a multi-scale fusion stage which has been proven to be profitable in minimizing the number of fragmented and overlapped segments. Line matching is performed by an iterative process that uses structural information collected through the use of different line neighborhoods, making the set of matched lines grow robustly at each iteration. Results show that this approach is suitable to deal with low-textured scenes, and also robust under a wide variety of image transformations. HighlightsA novel approach for line detection and matching is proposed..Lines are detected through a fusion process over the Gaussian scale-space.Lines are matched based on their appearance, geometric properties and neighborhoods.Our proposal minimizes the number of fragmented and overlapped segments.Our proposal is suitable for low-textured scenes, under uncontrolled illumination.

Decomposition of three-dimensional medical images into visual patterns

In this paper, we present a method for the decomposition of a volumetric image into its most relevant visual patterns , which we define as features associated to local energy maxima of the image. The method involves the clustering of a set of predefined bandpass energy filters according to their ability to segregate the different features in the image, thus generating a set of composite-feature detectors tuned to the specific visual patterns present in the data. Clustering is based on a measure of statistical dependence between pairs of frequency features. We will illustrate the applicability of the method to the initialization of a three-dimensional geodesic active model.