Personal Biometric Identification Based on ECG Features
O. Boumbarov, Y. Velchev, S. Sokolov
Key Words:
ECG signal; ECG wavelet; ECG segmentation; PCI feature extraction; ECG biometric personal identification.
Abstract:
In this paper, a new approach for personal biometric identification is presented based on extraction of ECG features. We perform denoising and segmentation on the input signal, after which we realize dimensionality reduction and select the features for recognition. The final stage of the proposed approach is classification and recognition of the extracted features.
3D Scenes Recovery through an Active Camera Based on Blur Assessment of the Resulting Image
K. Alexiev, I. Nikolova, G. Zapryanov
Key Words:
3D scene recovery; depth from defocus.
Abstract:
The paper considers the task of recovery of 3D information about the scene from single camera images. The basic idea is to extract the useful depth information from the images automatically and efficiently. Depth perception with single standard video surveillance camera is a challenging problem. The difficulties in deriving the distance to the observed objects in the scene can be partially overcome using active pan, tilt, zoom (PTZ) cameras and suitable control of camera parameters. There are several techniques for depth recovery. Here, the task of depth estimation in the context of the well known depth from defocus approach is considered. In this paper, it is proposed the problem to be solved using defocus blur. The characteristics of the approach are discussed. Experimental studies, using test patterns and real objects are presented.
Face Recognition Based on Linear Transforms and Classifier Combination
O. Boumbarov, K. Tonchev, S. Sokolov, S. Panev
Key Words:
Face recognition; dimensionality reduction; PCA; ICA; LDA; combined RBFNN classifiers.
Abstract:
This paper presents a novel combined approach to face recognition using subspace projections, based on PCA, ICA and LDA. First, initial dimensionality reduction and feature extraction with PCA is performed. After that, ICA is applied to extract non-correlated features for recognition. The next step is to apply LDA and improve feature class separability. The major impact for using LDA in FR is the representations in more discriminant space. Classification is performed with neural networks with radial basis functions. By using combined classifiers the overall reliability of the FR system is improved.
Face Detection Based on Skin Colour and Active Contours
À. Popov, D. Dimitrova, B. Naydenov
Key Words:
Fuzzy morphological operations; hit-or-miss transform; face detection; active contour.
Abstract:
This paper demonstrates a combined approach for finding a moving human face in a scene. First it considers the problem of finding facial features in frontal colour images. For a given image, our method first detects skin regions on it using colour segmentation in a 2D plane of selected colour space (HSV, YES, YCrCb or normalized RGB), and after that generates face candidates. Then, we apply methods of fuzzy mathematical morphology to obtain face candidates in order to find a particular feature a right eyebrow. After the face is outlined it is easy to initialize an active contour (snake) for tracing the same object in other frames of the scene. This research is essential due to the increasing security demands in the everyday life.
Pressure Evaluation in On-line and Off-line Signatures
D. Dimitrova, G. Gluhchev
Key Words:
Graphical tablet; scanner; signature; pressure evaluation; pressure distribution; sensor interoperability.
Abstract:
This paper presents a comparison of pressure of signatures acquired from a digitizing tablet using an inking pen. The written signature is digitized with a scanner and a pseudo pressure is evaluated. The experiments have shown that the obtained histograms are very similar and the two modalities could be used for pressure evaluation. Also, the data from the scanned image proved to be more stable which justifies its use for signature authentication.
Comparison of Two Approaches for Ultrasound Image Segmentation
P. Konstantinova, V. Behar, M. Nikolov
Key Words:
Image processing; quality measures; homomorphic transform; wavelet thresholding; binary images; image segmentation.
Abstract:
The image segmentation is often a desirable task which depends on image quality. The importance of ultrasound image quality in different application areas is indisputable. The image quality enhancement becomes an indispensable preprocessing step. In this paper the effect of some filtering procedures aimed to prepare images for segmentation, used as processing step, is studied. On the next step a segmentation algorithm, based on tracking algorithm in clutter using combination of the probabilistic data association approach and interacting multiple models approach, is used for contour determination. Another approach for contour determination, based on threshold technique producing binary image is also investigated. Both segmentation algorithms are tested on a simulated ultrasound image with known true positions of the contours and the evaluated errors are compared.