Catálogo de publicaciones - libros

Towards building more efficient human control interactive systems, we developed a neural network-based image processing system (called NEU-FACES), which first determines automatically whether or not there are any faces in given images and, if so, returns the location and extent of each face. Next, NEU-FACES uses neural network-based classifiers, which allow the classification of several facial expressions from features that we develop and describe. NEU-FACES is fully implemented and evaluated to assess its performance.

Iris identification () constitutes an increasingly accepted methodology of biometrics. is based on the successful encoding and matching of distinctive iris features (folds, freckles etc.), which - in turn - presupposes that iris segmentation has been accurately performed. In contrast to the inner (iris/pupil) iris boundary, which – owing to the high contrast between the adjacent areas - is relatively easy to localize, detection of the outer (iris/sclera) iris boundary is more challenging since the low contrast between the separated areas often results in fragmented, ambiguous and spurious edges. A novel approach to iris boundary detection is presented here, featuring a genetic algorithm (GA) for outer iris boundary detection.

In order to ensure that the performance of a fingerprint recognition system will be powerful with respect to the quality of input fingerprint images, the enhancement of fingerprints is essential. In this study wavelet transform and contourlet transform which is a new extension of the wavelet transform in two dimensions are applied for fingerprint enhancement. In addition, feature selection is a process that chooses a subset of features from the original fingerprint features so that the feature space is optimally reduced according to a certain criterion. In this study, a Genetic Algorithms (GAs) approach to fingerprint feature selection is proposed and selected features are input to Artificial Neural Networks (ANNs) for fingerprint recognition. The performance has been tested on fingerprint recognition.

In the paper a new way of intelligent medical pattern analysis directed for automatic semantic categorization and merit content understanding will be presented. Such an understanding will be based on the linguistic mechanisms of pattern interpretation and categorisation and is aimed at facilitation of in-depth analysis of the meaning for some classes of medical patterns, especially in the form of planar images or spatial reconstructions of selected organs. The approach presented in this paper will show the great possibilities of automatic lesion detection in the analysed structures using the grammar approach to the interpretation and classification tasks, based on cognitive resonance processes. Cognitive methods imitate the psychological and neurophysiological processes of understanding the analysed patterns or cases, as they take place in the brain of a qualified professional.

The performance of classifiers is commonly evaluated by classification rate and false alarm rate (FAR). Many applications like traffic monitoring, surveillance and other security relevant tasks suffer from the problem balancing the performance criteria in an appropriate way. In this contribution, we propose a kernel classification scheme with high performance in discriminating classes and rejecting clutter objects. Especially, it determines a class membership assessment. The classification scheme consists of two kernel classification stages and a maximum decision module as combiner. For tests, we use targets taken from the MSTAR synthetic aperture radar (SAR) dataset and clutter objects extracted from SAR scenes by a screening process. The dependency on parameter variations is shown and receiver operator characteristic (ROC) curves are given. The results confirm the high classification performance at low FARs. The integration into an operational demonstration system is in progress.

Automatic image compositing of very large data sets is necessary for the creation of extensive mosaics based on high spatial resolution remotely sensed imagery. A novel morphological image compositing algorithm has been developed which adapts to salient images edges. This technique produces seam lines that are difficult to identify by the naked eye which is also a characteristic to measure the quality of the resulting seam line. This paper begins with a description of the methodology and results based on Landsat 7 ETM+ imagery. It is also shown how updates to an already composited image data set can be easily made without having to reprocess the entire data set. Finally, ways of quantifying the quality of an automatically delineated cut line and future research directions are discussed.

In this paper the neural classifier for recognition of the circular shaped road signs is presented. This classifier belongs to the road signs recognition module, which in turn is a part of a driver assisting system. The circular shaped prohibition and obligation signs constitute the very important groups within the set of road signs. In this case however, it is not possible for a detector to determine rotation of the shapes that would allow dimension reduction of the search space. Thus the classifier has to be able to properly work with all possible affine deformations. To alleviate this problem we propose to use as features the statistical moments which were shown to be invariant within an affine group of transformations. The classification is performed by the probabilistic neural network which is trained with sign examples extracted from the real traffic scenes. The obtained results show good accuracy of classification and fast operation time.

In this paper, we present a novel classification method for eye location. It is based on image context analysis. There is general accord that context can be affluent derivation of information about an illumination, character and diversity of object. However, the problem of how to customize contextual influence is not yet solved clearly. Here we describe a naïve probabilistic method for modeling and testing the images of eye patterns. The proposed eye location method employs context-driven adaptive Bayesian framework to relive the effect due to uneven condition of face image. Based on an easy holistic analysis of face images, the proposed method is able to exactly locate eye position. The experimental results show that the proposed approach can achieve superior performance using various data sets to previously proposed methods.

This research explores the automated detection of surface blemishes in light-emitting diode (LED) chips. An LED is a semiconductor device that emits visible light when an electric current passes through the semiconductor chip. Water-drop blemishes, commonly appearing on the surfaces of chips, impair the appearance of LEDs as well as their functionality and security. Consequently, detecting water-drop blemishes becomes crucial for the quality control of LED products. We first use the one-level Haar wavelet transform to decompose a chip image and extract four wavelet characteristics. Then, the statistic of multivariate statistical analysis is applied to integrate the multiple wavelet characteristics. Finally, the wavelet based multivariate statistical approach judges the existence of water-drop blemishes. Experimental results show that the proposed method achieves an above 95% detection rate and a below 1.5% false alarm rate in inspecting water-drop blemishes of LED chips.

The increasing use of TFT-LCDs has generated a great deal of interest in manufacturing defects on TFT-LCD polarizing film because the poor quality of TFT-LCD polarizing film result in undesirable effects on the TFT-LCD display devices. In this paper, we propose a new inspection method that reliably detects various defects of TFT-LCD polarizing films. First, we apply a least mean squares adaptive filtering technique to remove background noise. Next, we use statistical characteristics to detect possible defects. Finally, we make a binary image to identify weather the TFT-LCD polarizing film has defects or not based on an adaptive threshold value. The performance of the proposed method has been evaluated on real TFT-LCD polarizing film samples.