Showing papers by "Tony Lindeberg published in 2003"

Real-time scale selection in hybrid multi-scale representations

Abstract: Local scale information extracted from visual data in a bottom-up manner constitutes an important cue for a large number of visual tasks. This article presents a framework for how the computation of such scale descriptors can be performed in real time on a standard computer. The proposed scale selection framework is expressed within a novel type of multi-scale representation, referred to as hybrid multi-scale representation, which aims at integrating and providing variable trade-offs between the relative advantages of pyramids and scale-space representation, in terms of computational efficiency and computational accuracy. Starting from binomial scale-space kernels of different widths, we describe a family pyramid representations, in which the regular pyramid concept and the regular scale-space representation constitute limiting cases. In particular, the steepness of the pyramid as well as the sampling density in the scale direction can be varied. It is shown how the definition of γ-normalized derivative operators underlying the automatic scale selection mechanism can be transferred from a regular scale-space to a hybrid pyramid, and two alternative definitions are studied in detail, referred to as variance normalization and lp-normalization. The computational accuracy of these two schemes is evaluated, and it is shown how the choice of sub-sampling rate provides a trade-off between the computational efficiency and the accuracy of the scale descriptors. Experimental evaluations are presented for both synthetic and real data. In a simplified form, this scale selection mechanism has been running for two years, in a real-time computer vision system.

Interest point detection and scale selection in space-time

Abstract: Several types of interest point detectors have been proposed for spatial images. This paper investigates how this notion can be generalised to the detection of interesting events in space-time data. Moreover, we develop a mechanism for spatio-temporal scale selection and detect events at scales corresponding to their extent in both space and time. To detect spatio-temporal events, we build on the idea of the Harris and Forstner interest point operators and detect regions in space-time where the image structures have significant local variations in both space and time. In this way, events that correspond to curved space-time structures are emphasised, while structures with locally constant motion are disregarded. To construct this operator, we start from a multi-scale windowed second moment matrix in space-time, and combine the determinant and the trace in a similar way as for the spatial Harris operator. All spacetime maxima of this operator are then adapted to characteristic scales by maximising a scale-normalised space-time Laplacian operator over both spatial scales and temporal scales. The motivation for performing temporal scale selection as a complement to previous approaches of spatial scale selection is to be able to robustly capture spatio-temporal events of different temporal extent. It is shown that the resulting approach is truly scale invariant with respect to both spatial scales and temporal scales. The proposed concept is tested on synthetic and real image sequences. It is shown that the operator responds to distinct and stable points in space-time that often correspond to interesting events. The potential applications of the method are discussed.

A Distance Measure and a Feature Likelihood Map Concept for Scale-Invariant Model Matching

Abstract: This paper presents two approaches for evaluating multi-scale feature-based object models. Within the first approach, a scale-invariant distance measure is proposed for comparing two image representations in terms of multi-scale features. Based on this measure, the maximisation of the likelihood of parameterised feature models allows for simultaneous model selection and parameter estimation. The idea of the second approach is to avoid an explicit feature extraction step and to evaluate models using a function defined directly from the image data. For this purpose, we propose the concept of a feature likelihood map, which is a function normalised to the interval [0, 1], and that approximates the likelihood of image features at all points in scale-space. To illustrate the applicability of both methods, we consider the area of hand gesture analysis and show how the proposed evaluation schemes can be integrated within a particle filtering approach for performing simultaneous tracking and recognition of hand models under variations in the position, orientation, size and posture of the hand. The experiments demonstrate the feasibility of the approach, and that real time performance can be obtained by pyramid implementations of the proposed concepts.

Fully automatic segmentation of MRI brain images using probabilistic anisotropic diffusion and multi-scale watersheds

Abstract: This article presents a fully automatic method for segmenting the brain from other tissue in a 3-D MR image of the human head. The method is a an extension and combination of previous techniques, and consists of the following processing steps: (i) After an initial intensity normalization, an affine alignment is performed to a standard anatomical space, where the unsegmented image can be compared to a segmented standard brain. (ii) Probabilistic diffusion, guided by probability measures between white matter, grey matter and cerebrospinal fluid, is performed in order to suppress the influence of extra-cerebral tissue. (iii) A multi-scale watershed segmentation step creates a slightly over-segmented image, where the brain contour constitutes a subset of the watershed boundaries. (iv) A segmentation of the over-segmented brain is then selected by using spatial information from the presegmented standard brain in combination with additional stages of probabilistic diffusion, morphological operations and thresholding. The composed algorithm has been evaluated on 50 T1-weighted MR volumes, by visual inspection and by computing quantitative measures of (i) the similarity between the segmented brain and a manual segmentation of the same brain, and (ii) the ratio of the volumetric difference between automatically and manually segmented brains relative to the volume of the manually segmented brain. The mean value of the similarity index was 0.9961 with standard deviation 0.0034 (worst value 0.9813, best 0.9998). The mean percentage volume error was 0.77 % with standard deviation 0.69 % (maximum percentage error 3.81 %, minimum percentage error 0.05 %).