Catálogo de publicaciones - libros

This paper describes a new approach for detection of curvilinear regions. These features detection can be useful for any matching based algorithm such as stereoscopic vision. Our detector is based on curvilinear structure model, defined observing the real world. Then, we propose a multi-scale search algorithm of curvilinear regions and we report some preliminary results.

Segmentation methods based on deformable models have proved to be successful with difficult images, particularly those using genetic algorithms to minimize the energy function. Nevertheless, they are normally conceived as fully automatic, and not always generate satisfactory results. In this work, a method to include the information of fixed points whithin a contour detection system using point distribution models and genetic algorithms is presented. Also, an interactive scheme is proposed to take advantage of this technique. The method has been tested against a database of 93 cattle images, with a significant improvement in the success rate of the detections, from 61% up to 95%.

We describe a model for tree crown extraction from aerial images, a problem of great practical importance for the forestry industry. The novelty lies in the prior model of the region occupied by tree crowns in the image, which is a phase field version of the higher-order active contour inflection point ‘gas of circles’ model. The model combines the strengths of the inflection point model with those of the phase field framework: it removes the ‘phantom circles’ produced by the original ‘gas of circles’ model, while executing two orders of magnitude faster than the contour-based inflection point model. The model has many other areas of application , to imagery in nanotechnology, biology, and physics.

This paper introduces a novel shape matching approach for the automatic identification of real world objects in complex scenes. The identification process is applied on isolated objects and requires the segmentation of the image into separate objects, followed by the extraction of representative shape features and the similarity estimation of pairs of objects. In order to enable an efficient object representation, a novel boundary-based shape descriptor is introduced, formed by a set of one dimensional signals called shape signatures. During identification, the cross-correlation metric is used in a novel fashion to gauge the degree of similarity between objects. The invariance of the method to uniform-scaling and partial occlusion is achieved by considering both cases as possible scenarios when correlating shape signatures. The proposed vision system is robust to ambient conditions (partial occlusion) and image transformations (scaling, rotation, translation). The performance of the identifier has been examined in a great range of complex image and prototype object selections.

Chazelle’s triangulation [1] forms today the common basis for linear-time Euclidean shortest path (ESP) calculations (where start and end point are given within a simple polygon). This paper provides an alternative method for subdividing a simple polygon into “basic shapes”, using trapezoids instead of triangles. The authors consider the presented method as being substantially simpler than the linear-time triangulation method. However, it requires a sorting step (of a subset of vertices of the given simple polygon); all the other subprocesses are linear time.

In this paper a new method for matching and recognition of shapes extracted from images, based on the structure and skeleton of the shape, is proposed. In this method, a function called “velocity function” derived from the radius function is introduced and values of this function are calculated for skeletal points. At this point, each skeletal curve is transformed into a vector containing values of the velocity function calculated along the curve. Then a tree-like structure is extracted from the skeleton of the shape so that each leaf of the tree indicates a skeletal curve of the skeleton and a velocity vector as its descriptor. These vectors are matched in fine-grain level using dynamic programming method and in coarse-grain level, the tree-like structures are matched using a greedy approach. At the final stage, the difference between the shapes is determined by a distance value resulted from the two levels matching process. This distance is used as a measure for shape matching and recognition. Experiments are performed on two standard binary image database in the presence of various transformations. Results confirm the efficiency and high recognition rate of our method.

Rigid 2D/3D image registration is a common strategy in medical image processing. In this paper we present an extended global optimization strategy for a rigid 2D/3D image registration that consists of three components: a combination of a global and a local optimizer, a combination of a multi-scale and a multi-resolution approach, and a combination of an in-plane and an out-of-plane registration. The global optimizer Adaptive Random Search is used to provide several coarse registration results on a low resolution level that are refined by the local optimizer Best Neighbor on a higher resolution level.

We evaluate the performance and the precision of our registration algorithm using two phantom models. We could approve that all three components of our optimization strategy lead to an significant improvement of the registration.

Recently, the concept of consistent image registration has been introduced to refer to a set of algorithms that estimate both the direct and inverse deformation together, that is, they exchange the roles of the target and the scene images alternatively; it has been demonstrated that this technique improves the registration accuracy, and that the biological significance of the obtained deformations is also improved.

When dealing with free form deformations, the inversion of the transformations obtained becomes computationally intensive. In this paper, we suggest the parametrization of such deformations by means of a cubic B-spline, and its approximated inversion using a highly efficient algorithm. The results show that the consistency constraint notably improves the registration accuracy, especially in cases of a heavy initial misregistration, with very little computational overload.

Although interfering (outlying) details complicate image re- cognition and retrieval, ‘soft masking’ of outliers shows considerable promise for robust pixel-by-pixel image matching or reconstruction from principal components (PC). Modeling the differences between two images or between an image and its PC estimate (obtained as a projection onto a subspace of PCs) with a mixed distribution of random noise and outliers, the masks are produced by a simple iterative Expectation-Maximisation based procedure. Experiments with facial images (extracted from the MIT face database) demonstrate the efficiency of this approach.

The aim of this work is to code the string matching problem as an optimization task and carrying out this optimization problem by means of a Hopfield neural network. The proposed method uses TCNN, a Hopfield neural network with decaying self-feedback, to find the best-matching (i.e., the lowest global distance) path between an input and a template. The proposed method is more than ‘exact’ string matching. For example wild character matches as well as character that never match may be used in either string. As well it can compute edit distance between the two strings. It shows a very good performance in various string matching tasks.