Showing papers by "Ioannis Pitas published in 1997"

Rule-based face detection in frontal views

Abstract: Face detection is a key problem in building automated systems that perform face recognition A very attractive approach for face detection is based on multiresolution images (also known as mosaic images) Motivated by the simplicity of this approach, a rule-based face detection algorithm in frontal views is developed that extends the work of G Yang and TS Huang (see Pattern Recognition, vol27, no1, p53-63, 1994) The proposed algorithm has been applied to frontal views extracted from the European ACTS M2VTS database that contains the videosequences of 37 different persons It has been found that the algorithm provides a correct facial candidate in all cases However, the success rate of the detected facial features (eg eyebrows/eyes, nostrils/nose, and mouth) that validate the choice of a facial candidate is found to be 865% under the most strict evaluation conditions

Embedding robust watermarks by chaotic mixing

Abstract: This paper presents a watermarking algorithm for copyright protection of digital images. A copyright label represented by a binary image is embedded in grayscale or color digital image. A mixing dynamical system controls embedding, detection and reconstruction of the copyright label. Detection of the watermark is succeeded either by direct reconstruction of the watermark as a binary image or by using statistical hypothesis testing.

A Fully Automatic Approach to Facial Feature Detection and Tracking

Abstract: The detection of facial features is a necessary step for a wide range of applications e.g. person verification, lipreading and model-based face coding. Due to changes in illumination, visual angle and facial expressions, the variability of facial features in their appearance is high. A robust approach to facial feature detection has to handle these variations. In this framework, we present an approach for the extraction of eyebrows, eyes, nostrils, mouth and chin. Our approach for facial feature extraction is based on the observation that facial features differ from the rest of the face because of their low brightness. Thus, we detect facial feature candidates by evaluating the topographic greylevel relief of the face region. Based on vertical symmetry, distances between facial features and the assessment of each facial feature, we choose the best face constellation. Incomplete face constellations are considered as well. Once facial features are detected in an image sequence, they can be tracked over time. We perform facial feature tracking by block matching. The best-match position is refined by minima analysis. The success of our approach was tested on 38 different image sequences containing faces.

Computerized 3‐D reconstruction of two “double teeth”

Abstract: "Double teeth" is a root malformation in the dentition and the purpose of this study was to reconstruct three-dimensionally the external and internal morphology of two "double teeth". The first set of "double teeth" was formed by the conjunction of a mandibular molar and a premolar, and the second by a conjunction of a maxillary molar and a supernumerary tooth. The process of 3-D reconstruction included serial cross-sectioning, photographs of the sections, digitization of the photographs, extraction of the boundaries of interest for each section, surface representation using triangulation and, finally, surface rendering using photorealistic effects. The resulting three-dimensional representations of the two teeth helped us visualize their external and internal anatomy. The results showed: a) in the first case, fusion of the radical and coronal dentin, as well as fusion of the pulp chambers; and b) in the second case, fusion only of the radical dentin and the pulp chambers.

On fast running max-min filtering

Abstract: The problem of fast running max/min filters for arbitrary size windows is addressed. The size of the filter window is increased to the least power of two greater than the given size and, the input sequence is expanded. The running max/min computation uses a fast algorithm for power of two window sizes. The computational complexity (comparisons per sample) of the proposed algorithm is very close to log/sub 2/ n, where n is the size of the given window. A flexible hardware implementation for n ranging between two consecutive powers of two is discussed.

Face authentication based on morphological grid matching

Abstract: A novel dynamic link architecture based on multiscale morphological dilation-erosion is proposed for face verification in a cooperative scenario where the candidates claim an identity that is to be checked. The performance of the morphological dynamic link architecture (MDLA) is evaluated in terms of the receiver operating characteristic (ROC) for several threshold selections on the matching error in the M2VTS database. The experimental results indicate that the proposed method outperforms the dynamic link matching with Gabor based feature vectors.

Face Authentication Using Morphological Dynamic Link Architecture

Abstract: A very attractive approach for face detection is based on multiresolution images (also known as mosaic images). Motivated by the simplicity of this approach, a rule-based face detection algorithm in frontal views is developed first. Second, a novel dynamic link architecture based on multiscale morphological dilation-erosion is proposed for face authentication. More specifically, a sparse grid is placed over the outcome of face detection stage for each person in a reference set. Subsequently, multiscale morphological operations are employed to yield a feature vector at each node of the grid and dynamic link matching is applied to verify the identity of each person from a test set. The first experimental results reported in this paper verify the superiority of the proposed method over the (standard) dynamic link matching that is based on Gabor wavelets.

Facial feature extraction using adaptive Hough transform, template matching and active contour models

Abstract: The present paper describes an extension to the methods proposed by Sobottka and Pitas (see Proc. of the IEEE Int. Conf. on Image Processing, Lausanne, Switzerland, p.483-6, 1996) for the extraction of facial features with the ultimate goal to be used in defining a sufficient set of distances between them so that a unique description of the structure of a face is obtained. Eyebrows, eyes, nostrils, mouth, cheeks and chin are considered as interesting features. Candidate for eyes, nostrils, mouth are determined by searching for minima and maxima on the x- and y-projections of the grey-level relief. Candidates for cheeks and chin are determined by performing adaptive Hough transform on a relevant subimage defined according to the position of the eyes and mouth and the ellipse containing the main connected components of the image. A deforming technique is also applied to the ellipse representing the main face region, in order to acquire a more accurate model of the face. Candidates for eyebrows are determined by adapting a proper grey-level template to an area restricted by the position of the eyes. The algorithms presented were tested on a set of 37 different color images containing features such as beard, glasses and changing facial expressions.

An error concealment scheme for MPEG-2 coded video sequences

Abstract: The problem of errors occurring in MPEG-2 coded video sequences, caused by signal loss during transmission, is examined in this paper and an attempt is made to reconstruct the lost parts at each frame. The proposed error concealment scheme exploits reconstructed temporal information from previously decoded frames in order to conceal bitstream errors in all types of frames: I, P, or B, as long as temporal information is available. Since no such information is available for the first frame (I-frame) of an MPEG-2 coded sequence, another concealment technique is added to the proposed scheme, which uses spatial information from neighbouring macroblocks (MBs). The simulation results compared with other methods prove to be better judging from both PSNR values and the perceived visual quality of the reconstructed sequence. Its quality ameliorates with time.

Morphological Iterative Closest Point Algorithm

Abstract: This paper describes a method for accurate and computationally efficient registration of 3-D shapes including curves and surfaces. The method is based on the iterative closest point (ICP) algorithm. The real strength of our algorithm is the use of Morphological Voronoi tessellation method to construct the Voronoi regions around the seed points with respect to a certain distance metric. The tessellated domain can be used to avoid the most computationally expensive step in the ICP, namely to find the closest points Thus the proposed algorithm is much faster than classical ICP.

Randomized fuzzy cell Hough transform

Abstract: Randomized Hough transform (RHT) has been recently proposed as a new and efficient variation of the Hough transform for curve detection. In this paper the RHT is combined with the fuzzy cell Hough transform (FCHT) and a new variation, the randomized fuzzy cell Hough transform (RFCHT) is proposed. The p-dimensional parameter space of Hough transform is split into fuzzy cells with overlapped intervals of confidence. The fuzzy cells are defined as fuzzy numbers. The RFCHT selects p pixels from an edge image by random sampling and solves the p parameters of a curve. Then the p parameters accumulate to more than one fuzzy cells, since the fuzzy cells intervals are overlapped by adding a value that belongs in the interval [0, 1] and is calculated from the membership function of the corresponding fuzzy cell. The procedure continues by using a percentage of the total contour pixels. The RFCHT algorithm preserves the advantages of RHT and FCHT. The algorithm has good computational speed and small storage requirements due to random sampling and correct and more accurate detections, especially in noisy images, due to fuzzy cells.

Perspective distortion analysis for mosaicing images painted on cylindrical surfaces

Abstract: A set of monocular images of a curved painting is taken from different viewpoints around its curved surface. After deriving the surface localization in the camera coordinate system we backproject the image on the curved surface and we flatten it. We analyze the perspective distortions of the scene in the case when it is mapped on a cylindrical surface. Based on the result of this analysis we derive the necessary number of views in order to represent the entire scene depicted on a cylindrical surface. We employ a matching-based mosaicing method for reconstructing the scene from the curved surface. The proposed method is appropriate to be used for painting reconstruction.

Face authentication based on morphological shape decomposition

Abstract: A novel morphological dynamic link architecture that employs morphological shape decomposition as a feature extraction mechanism has been developed and tested. The first experimental results collected are very encouraging and indicate that the proposed method outperforms the (standard) dynamic link matching that is based on Gabor wavelets.

Mosaicing of Flattened Images from Straight Homogeneous Generalized Cylinders

Abstract: This paper presents a new method for reconstructing paintings from component images. A set of monocular images of a painting from a straight homogeneous generalized cylinder is taken from various viewpoints. After deriving the surface localization in the camera, coordinate system, the images are backprojected on the curved surface and flattened. We derive the perspective distortion of the scene in the case when it is mapped on a cylindrical surface. Based on the result of this study we derive the necessary number of views in order to represent the entire scene depicted on a cylindrical surface. We propose a matching-based mosaicing algorithm for reconstructing the scene from the curved surface. The proposed algorithm is applyed on paintings.

Moving scene segmentation using median radial basis function network

Abstract: Various approaches were suggested for simultaneous optical flow estimation and segmentation in image sequences. In this study, the moving scene is decomposed in different regions with respect to their motion, by means of a pattern recognition scheme. The inputs of the proposed scheme are the feature vectors representing still image and motion information. The classifier employed is the Median Radial Basis Function (MRBF) neural network. Each class corresponds to a moving object. An error criterion function derived from the probability estimation theory and related to the moving scene model is used as cost function. Marginal median and median of the absolute deviations estimators are employed for estimating the basis function parameters.

Application of Neural Networks and Order Statistics Filters to Speckle Noise Reduction in Remote Sensing Imaging

Abstract: A novel approach to suppression of speckle noise in remote sensing imaging based on a combination of segmentation and optimum L-filtering is presented. With the aid of a suitable modification of the Learning Vector Quantizer (LVQ) neural network, the image is segmented in regions of (approximately) homogeneous statistics. For each of the regions a minimum mean-squared-error (MMSE) L-filter is designed, by using the histogram of grey levels as an estimate of the parent distribution of the noisy observations and a suitable estimate of the (assumed constant) original signal in the corresponding region. Thus, a bank of L-filters results, with each of them corresponding to and operating on a different image region. Simulation results are presented, which verify the (qualitative and quantitative) superiority of our technique over a number of commonly used speckle filters.

Mosaïque de peintures sur cylindres généralisés uniformes à axe droit

Abstract: Le problème traité concerne la reconstruction de scènes présentes sur des supports modélisés par des cylindres généralisés uniformes à axe droit. Pour cela il est nécessaire de localiser les vues, de les reprojeter afin d’obtenir des images redressées. Nous effectuons alors une mosaïque d’images reconstituant la scène complète. L’originalité du travail est son application aux peintures sur voûte ou sur colonne considérées comme étant des cylindres généralisés uniformes à axe droit. Ces travaux, effectués en vision monoculaire, utilisent des connaissances a priori sur les surfaces étudiées. Les résultats nouveaux sont la localisation à partir d’une seule vue et la mise en correspondance automatique dans l’algorithme de mosaïque. ABSTRACT

Constrained adaptive multichannel L-filters

Abstract: Adaptive multichannel L-filters based on marginal ordering when structural constraints such as location invariance or unbiasedness are imposed on the filter coefficients are proposed. Two novel adaptive algorithms are derived by using the Frost (1972) algorithm for minimizing the mean squared error subject to constraints in LMS and in LMS-Newton algorithms. It is demonstrated by experiments that the Frost-LMS algorithm has a faster convergence rate than the Frost LMS-Newton algorithm but it yields a higher steady-state MSE.