Many different low-level feature detectors exist and it is widely agreed that the evaluation of detectors is important. In this paper we introduce two evaluation criteria for interest points: repeatability rate and information content. Repeatability rate evaluates the geometric stability under different transformations. Information content measures the distinctiveness of features. Different interest point detectors are compared using these two criteria. We determine which detector gives the best results and show that it satisfies the criteria well.

. The appearance of an object is composed of local structure. This local structure can be described and characterized by a vector of local features measured by local operators such as Gaussian derivatives or Gabor filters. This article presents a technique where appearances of objects are represented by the joint statistics of such local neighborhood operators. As such, this represents a new class of appearance based techniques for computer vision. Based on joint statistics, the paper develops techniques for the identification of multiple objects at arbitrary positions and orientations in a cluttered scene. Experiments show that these techniques can identify over 100 objects in the presence of major occlusions. Most remarkably, the techniques have low complexity and therefore run in real-time. 1. Introduction The paper proposes a framework for the statistical representation of the appearance of arbitrary 3D objects. This representation consists of a probability density function or jo...

We present a novel geometric approach for solving the stereo problem for an arbitrary number of images (greater than or equal to 2). It is based upon the denition of a variational principle that must be satisfied by the surfaces of the objects in the scene and their images. The Euler-Lagrange equations which are deduced from the variational principle provide a set of PDE's which are used to deform an initial set of surfaces which then move towards the objects to be detected. The level set implementation of these PDE's potentially provides an efficient and robust way of achieving the surface evolution and to deal automatically with changes in the surface topology during the deformation, i.e. to deal with multiple objects. Results of an implementation of our theory also dealing with occlusion and vibility are presented on synthetic and real images.

Nonlinear diffusion filtering is usually performed with explicit schemes. They are only stable for very small time steps, which leads to poor efficiency and limits their practical use. Based on a recent discrete nonlinear diffusion scale-space framework we present semi-implicit schemes which are stable for all time steps. These novel schemes use an additive operator splitting (AOS) which guarantees equal treatment of all coordinate axes. They can be implemented easily in arbitrary dimensions, have good rotational invariance and reveal a computational complexity and memory requirement which is linear in the number of pixels. Examples demonstrate that, under typical accuracy requirements, AOS schemes are at least ten times more efficient than the widely-used explicit schemes.

We present a novel geometric approach for solving the stereo problem for an arbitrary number of images (greater than or equal to 2). It is based upon the denition of a variational principle that must be satised by the surfaces of the objects in the scene and their images. The Euler-Lagrange equations which are deduced from the variational principle provide a set of PDE's which are used to deform an initial set of surfaces which then move towards the objects to be detected. The level set implementation of these PDE's potentially provides an efficient and robust way of achieving the surface evolution and to deal automatically with changes in the surface topology during the deformation, i.e. to deal with multiple objects. Results of an implementation of our theory also dealing with occlusion and vibility are presented on synthetic and real images.

Many computer vision tasks rely on feature extraction. Interest points are such features. This paper shows that interest points are geometrically stable under different transformations and have high information content (distinctiveness). These two properties make interest points very successful in the context of image matching. To measure these two properties quantitatively, we introduce two evaluation criteria : repeatability rate and information content. The quality of the interest points depends on the detector used. In this paper several detectors are compared according to the criteria specified above. We determine which detector gives the best results and show that it satisfies the criteria well. 1 Introduction Interest points are locations in the image where the signal changes two-dimensionally. Examples include corners 1 and T-junctions, as well as locations where the texture varies significantly. Figure 1 shows an example of interest points detected on the "sower" painting...

We are considering the problem of recovering the three-dimensional geometry of a scene from binocu-lor stereo disparity. Once a dense disparity map has been computed from a stereo pair of images, one of-ten needs to calculate some local diferential proper-ties of the cowesponding 3-0 surface such as orien-tation or curvatures. The wual approach is to build a 3-0 reconstruction of the surface(s) from which all shape properties will then be derived without ever go-ing back to the original images. In this paper, we de-part from this paradigm and propose to w e the images directly to compute the shape properties. We thus pro-pose a new method extending the classical cowelation method to estimate accurately both the disparity and its derivatives directly from the image data. We then relate those derivatives to diferential properties of the surface such as orientation and curvatures.

We present a new image-based technique to rigidly register intra-operative 3D ultrasound (US) with pre-operative Magnetic Resonance (MR) images. Automatic registration is achieved by maximization of a similarity measure which generalizes the correlation ratio (CR), and whose novelty is to incorporate multivariate information from the MR data (intensity and gradient). In addition, the similarity measure is built upon a robust intensity-based distance measure, which makes it possible to handle a variety of US artifacts. A cross-validation study has been carried out using a number of phantom and clinical data. This indicates that the method is quite robust and that the worst registration errors are of the order of the MR image resolution. Keywords: image registration, ultrasound, magnetic resonance, correlation ratio, robust estimation. 1

Most algorithms in 3D computer vision rely on the pinhole camera model because of its simplicity, whereas video optics, especially low-cost wide-angle or fish-eye lenses, generate a lot of non-linear distortion which can be critical. To find the distortion parameters of a camera, we use the following fundamental property: a camera follows the pinhole model if and only if the projection of every line in space onto the camera is a line. Consequently, if we find the transformation on the video image so that every line in space is viewed in the transformed image as a line, then we know how to remove the distortion from the image. The algorithm consists of first doing edge extraction on a possibly distorted video sequence, then doing polygonal approximation with a large tolerance on these edges to extract possible lines from the sequence, and then finding the parameters of our distortion model that best transform these edges to segments. Results are presented on real video images, compared with distortion calibration obtained by a full camera calibration method which uses a calibration grid.

We propose a new strategy to design recursive implementations of the Gaussian filter and Gaussian regularized derivative filters. Each recursive filter consists of a cascade of two stable N -order subsystems (causal and anti-causal). The computational complexity is 2N multiplications per pixel per dimension independent of the size (s) of the Gaussian kernel. The filter coefficients have a closed-form solution as a function of scale (s) and recursion order N (N=3,4,5). The recursive filters yield a high accuracy and excellent isotropy in n-D space.