Catálogo de publicaciones - libros

This paper presents a new contribution for the problem of automatic visual inspection. New methods for determining threshold values for fabric defect detection using feedforward neural networks are proposed. Neural networks are one of the fastest most flexible classification systems in use. Their implementation in defect detection, where a clear classification is needed, requires thresholding the output. Two methods are proposed for threshold selection, statistical analysis of the NN output and confusion matrix based optimization. Experimental results obtained from the real fabric defects, for the two approaches proposed in this paper, have confirmed their usefulness.

A method for pedestrian detection from real world outdoor scenes is presented in this paper. The technique uses disparity information, ground plane estimation and biometric information based on the golden ratio. It can detect pedestrians even in the presence of severe occlusion or a lack of reliable disparity data. It also makes reliable choices in ambiguous areas since the pedestrian regions are initiated using the disparity of head regions. These are usually highly textured and unoccluded, and therefore more reliable in a disparity image than homogeneous or occluded regions.

This paper describes an image processing system developed for automatic counting the number of collembola individuals on petri disks images. The system uses image segmentation and mathematical morphology techniques to identify and count the number of collembolans. The main challenges are the specular reflections at the edges of the circular samples and the foam present in a number of samples. The specular reflections are efficiently identified and removed by performing a two-stage segmentation. The foam is considered to be noise, as it is at cases difficult to discriminate between the foam and the collembola individuals. Morphological image processing tools are used both for noise reduction and for the identification of the collembolans. A total of 38 samples (divided in 3 groups according to their noise level) were tested and the results produced from the automatic system compared to the values available from manual counting. The relative error was on average 5.0% (3.4% for good quality samples, 4.6% for medium quality and 7.5% for poor quality samples).

This paper present a method for extracting and automatic tracking of human body using template matching and human body model fitting for specific activity. The method includes training and testing stages. For training, the body shapes are manually segmented from image sequences as templates and are clustered. The 2D joint locations of each cluster center are labeled and the dynamical models of the templates are learned. For testing, a “seed” frame is first selected from the sequence according to the reliability of motion segmentation and several most matched templates to it are obtained. Then, a template tracking process within a probabilistic framework integrating the learnt dynamical model is started forwards and afterwards until the entire sequence is matched. Thirdly, a articulated 2D human body model is initialized from the matched template and then iteratively fit to the image features. Thus, the human body segmentation results and 2D body joints are got. Experiments are performed on broadcasted diving sequences and promising results are obtained. We also demonstrate two applications of the proposed method for sports training.

Nowadays, the heterogeneity of networks, terminals, and users is growing. At the same time, the availability and usage of multimedia content is increasing, which has raised the relevance of content adaptation technologies able to fulfill the needs associated to all usage conditions. For example, mobile displays tend to be too small to allow one to see all the details of an image. This paper presents an innovative method to integrate low-level and semantic visual cues into a unique visual attention map that represents the most interesting contents of an image, allowing the creation of a video sequence that browses through the image displaying its regions of interest in detail. The architecture of the developed adaptation system, the processing solutions and also the principles and reasoning behind the algorithms that have been developed and implemented are presented in this paper. Special emphasis is given to the integration of low-level and semantic visual cues for the maximization of the image to video adapted experience.

The problem of determining the script and language of a document image has a number of important applications in the field of document analysis, such as indexing and sorting of large collections of such images, or as a precursor to optical character recognition (OCR). In this paper, we investigate the use of texture as a tool for determining the script of a document image, based on the observation that text has a distinct visual texture. An experimental evaluation of a number of commonly used texture features is conducted on a newly created script database, providing a qualitative measure of which features are most appropriate for this task. Strategies for improving classification results in situations with limited training data and multiple font types are also proposed.

The study of biology and population dynamics of fish species requires the estimation of fecundity in individual fish in a routine way in many fisheries laboratories. The traditional procedure used by fisheries research is to count the oocytes manually on a subsample of known weight of the ovary, and to measure few oocytes under a binocular microscope. This process could be done on a computer using an interactive tool to count and measure oocytes. In both cases, the task is very time consuming, which implies that fecundity studies are rarely conducted routinely. This work represents the first attempt to design an automatic algorithm to recognize the oocytes in histological images. Two approaches based on region and edge information are described to segment the image and extract the oocytes. An statistical analysis reveals that higher than 74% of oocytes are recognized for both approaches, when an overlapping area between machine detection and true oocyte demanded is greater than 75%.

In this paper, we propose a new method for the indexing and classification of Islamic Stars Pattern (ISP) images based on rotational symmetry information. A computational model for the extraction of rotational symmetry features is proposed. This model is based on the three following steps. First, we detect the rotation center of the ISP image, then we complete the image structure by using symmetry information. Second, we compute the angle of rotation and number of folds. Finally, we extract the fundamental region, a representative region in the image from which the whole image can be regenerated. A method is also proposed for indexing and classifying ISP images on the basis of the extracted features. The classification algorithm is based on the number of folds. We characterize an image by its fundamental region and by its class which is defined in the classification step. Experiments show promising results either for ISP images classification or indexing. Efforts for the subsequent task of repeated pattern images classification can be significantly reduced.

Automatic evaluation of visual print quality is addressed in this study. Due to many complex factors of perceived visual quality its evaluation is divided to separate parts which can be individually evaluated using standardized assessments. Most of the assessments however require active evaluation by trained experts. In this paper one quality assessment, missing dot detection from printed dot patterns, is addressed by defining sufficient hardware for image acquisition and method for detecting and counting missing dots from a test strip. The experimental results are evidence how the human assessment can be automated with the help of machine vision, thus making the test more repeatable and accurate.

BigBatch is a processing environment designed to automatically process batches of millions of monochromatic images of documents generated by production line scanners. It removes noisy borders, checks and corrects orientation, calculates and compensates the skew angle, crops the image standardizing document sizes, and finally compresses it according to user defined file format. BigBatch encompasses the best and recently developed algorithms for such kind of document images. BigBatch may work either in standalone or operator assisted modes. Besides that, BigBatch in standalone mode is able to process in clusters of workstations or in grids.