Catálogo de publicaciones - libros

We propose a novel method for measuring a whole three-dimensional shape of an object with a simple structured system. The proposed system consists of a single CCD camera and a cylindrical mirror. A target object is placed inside the cylindrical mirror and an image is captured by the camera from above. There are two or more points that have the same origin in the captured image: one is observed directly, and the other is observed via the mirror. This situation means that a point is observed by a real camera and virtual cameras at the same time. Therefore, the three-dimensional shape of the object can be obtained using stereo vision. We simulated an actual experimental situation and measured the three-dimensional shape of the object from the simulated image, and the results have demonstrated that the proposed method is useful for measuring the whole three-dimensional shape.

Mottling is one of the most important printing defects in modern offset printing using coated papers. Mottling can be defined as undesired unevenness in perceived print density. In our research, we have implemented three methods to evaluate print mottle: the standard method, the cluster-based method, and the bandpass method. Our goal was to study the methods presented in literature, and modify them by taking relevant characteristics of the human visual system into account. For comparisons, we used a test set of 20 grey mottle samples which were assessed by both humans and the modified methods. The results show that when assessing low-contrast unevenness of print, humans have diverse opinions about quality, and none of the methods accurately capture the characteristics of human vision.

In situ detection and identification of microorganisms in the environment are important in general microbial ecology. Also the rapid inspection of microbial contamination at food processing plant is urgent task. We propose a method of detecting and identifying microorganisms for rapid inspection using spectral imaging technique. Spectral images of photosynthetic and non-photosynthetic bacterial colonies having different absorption spectra in near infrared wavelength region were measured directly from Petri dish. Bacterial region in the images was first detected and then identified using multiple discriminant analysis. Detection and identification errors for various sized colonies were analyzed. As the result, colonies with diameters of 100 and 300 m were detected and identified with sufficient accuracy, respectively. This means the time for detection and identification can be shorten less than a half and about several weeks compared with the conventional methods.

When studying animal behaviour and ecology the recognition of individuals is very important and in the case of bottlenose dolphins this can be done via photo-identification of their dorsal fins. Here we develop a mathematical model that describes this fin shape which is then fitted to the data by a Bayesian approach implemented using MCMC methods. This project is still at a testing stage and we are currently working with simulated data. Future work includes: extracting the outline of the fin shape from the pictures; fitting the model to real data; and devising a way of using the model to discriminate between individuals.

We develop a statistical shape model for the analysis of shape variation. In particular, we consider models of shapes that exhibit self-similarity along their contours such as fractal and space filling curves. Overlapping contour segments are parametrically modelled using an orthogonal basis set, Legendre Polynomials, and used to estimate similarity transformations to a reference segment, which may or may not be from the contour being analysed. The alignment is affine and regresses the model to the data by least squares fitting and is followed by a PCA of the coregistered set of contour segments. The local shape space is defined jointly by the segment-to-segment ‘warps’ and the mean plus eigen vectors of the shape space, hence Warplets. The parametric modelling makes the alignment correspondence-free so that arbitrary sized segments can be aligned and the local warps can be inverted to reconstruct model approximations of the data. The approach shows potential in capturing fine details of shape variation and is applicable to complex shapes and those with repetitive structure, when only a few training examples are available.

In this paper we study how learning can be used in several aspects of car detection and tracking. The overall goal is to develop a system that learns its surrounding and subsequently does a good job in detecting and tracking all cars (and later pedestrians and bicycles) in an intersection. Such data can then be analyzed in order to determine how safe an intersection is. The system is designed to, with minimal supervision, learn the location of the roads, the geometry needed for rectification, the size of the vehicles and the tracks used to pass the intersection. Several steps in the tracking process are described. The system is verified with experimental data, with promising results.

The recognition of painted strokes is an important step in analyzing underdrawings in infrared reflectogramms. But even for art experts, it is difficult to recognize all drawing tools and materials used for the creation of the strokes. Thus the use of computer-aided imaging technologies brings a new and objective analysis and assists the art experts. This work proposes a method to recognize strokes drawn by different drawing tools and materials. The method uses texture analysis algorithms performing along the drawing trace to distinguish between different types of strokes. The benefit of this method is the increased content of textural information within the stroke and simultaneously in the border region. We tested our algorithms on a set of six different types of strokes: 3 classes of fluid and 3 classes of dry drawing materials.

Computerized analysis of mammograms can serve as a secondary opinion, improving consistency by providing a standardized approach to mammogram interpretation, and increasing detection sensitivity. However, before any computer aided mammography algorithm can be applied to the nipple, one of several important anatomical features, need to extracted. This is challenging since the contrast near the border of the breast, and thus the nipple in mammograms, is very low. Therefore, in order to develop more robust and accurate methods, it is important to restrict the search area for automatic nipple location. We propose an automatic initialization of the search area of the nipple by combining a geometrical assumptions verified against the MIAS database regarding the location of the nipple along the breast border and a geometrical model for deciding how far into the breast region the nipple can occur. In addition, the modelling reduces the need for parameters determining the search area and thus making the method more general. We also investigate the variance between the medical experts often use as ground truth when determining performance measures for developed medical methods.

Centerline extraction of tubular structures such as blood vessels and airways in 3D volume data is of vital interest for applications involving registration, segmentation and surgical planing. In this paper, we propose a robust method for 3D centerline extraction of tubular structures. The method is based on a novel multiscale medialness function and additionally provides an accurate estimate of tubular radius. In contrast to other approaches, the method does not need any user selected thresholds and provides a high degree of robustness. For comparison and performance evaluation, we are using both synthetic images from a public database and a liver CT data set. Results show the advantages of the proposed method compared with the methods of Frangi et al. and Krissian et al.

The channel representation allows the construction of soft histograms, where peaks can be detected with a much higher accuracy than in regular hard-binned histograms. This is critical in e.g. reducing the number of bins of generalized Hough transform methods. When applying the maximum entropy method to the channel representation, a minimum-information reconstruction of the underlying continuous probability distribution is obtained.

The maximum entropy reconstruction is compared to simpler linear methods in some simulated situations. Experimental results show that mode estimation of the maximum entropy reconstruction outperforms the linear methods in terms of quantization error and discrimination threshold. Finding the maximum entropy reconstruction is however computationally more expensive.