Catálogo de publicaciones - libros

Among gestures naturally performed by users during communication, pointing gestures can be easily recognized and included in more natural new Human Computer Interfaces. We approximate the eye-finger pointing direction of a user by detecting and tracking, in real time, the 3D positions of the centre of the face and of both hands; the positions are obtained by a stereoscopic device located on the top of the display. From the head position and biometric constraints, we define both a rest area and an action area. In this former area, the hands are searched for and the pointing intention is detected. The first hand spontaneously moved forward by the user is defined as the pointing hand whereas the second detected hand, when it first moves forwards, is considered as the selection hand. Experiments on spatial precision, carried out with a group of users, show that the minimum size of an object to be easily pointed at is some 1.5 percent of the diagonal of the large display.

Omnidirectional video is an adequate image-based scene representation data format for interactive walkthroughs or surround viewing of a scene over the Internet. While efficient compression is a key issue in this context, the properties of omnidirectional video are not in line with the assumptions that are made about video sequences by modern compression standards like MPEG-4 AVC/H.264. We introduce a preprocessing approach which transforms omnidirectional video into a sequence of panoramic images before encoding. Using a state of the art MPEG-4 AVC/H.264 video coder, our approach performs up to 2dB better for low bit rates compared to regular encoding of omnidirectional video.

In this paper, we present a compact survey of bit-rate control techniques which can be used in design of video codecs equipped with online information on dynamical ROIs. We introduce the concept of zero rate increment on level sets in the context of linear data source models in ñ-domain. This approach enables a smooth image fidelity degradation on level sets for distance function from pixel to ROI.We focus in our examples on H.264/AVC which is the state-of-the-art in the video codec area.

In the existing inter-frame wavelet video coding techniques, the poorly matched pixels from adjacent frames are temporally filtered into the high-pass temporal frame in the same way as the good matches using motioncompensated prediction (MCP). This tends to result in high-energy blocks being introduced into the high-pass frame, which leads to less efficient spatial coding of the affected frames. We propose to use prediction from spatial neighbouring pixels for high-pass filtering of the poorly matched blocks, which results in a significantly improved performance at low bitrates with no loss at high bitrates.

Video downscaling provides an effective content adaptation method to support end-to-end delivery of video over heterogeneous networks. This paper presents a discrete cosine transform (DCT)-domain downscaling algorithm for MPEG-2 video. The test results demonstrate the efficiency of this method in terms of downscaled video quality and computational complexity.

This paper focuses on generalized weighted vector filters which constitute a powerful filtering frameworkfor multichannel signal processing. The utilization of a generalized multichannel sigmoidal adaptation scheme allows to accommodate the filter parameters to varying signal and noise statistics. Experimental results indicate that the presented multichannel filters are capable of removing impulsive noise present in color images and simultaneously they preserve color and structural information.

We describe in detail a method for calibrating a multispectral imaging system based on interference filters. The calibration aims to remove introduced by the sensor, and optic and/or filters from multispectral images. After which, we can correct the non-linearity of the sensor response. Systematic noises are measured through a rigorous protocol for acquiring offset, and thermal, and Flat-Field images. The methods for acquiring Flat-Field image, and linearizing sensor response are novel and particularly efficient in the case of a multispectral imaging system. Indeed, in such a system, the reconstruction of a spectrum for each pixel comes from the set of values taken by this pixel along the sequence of channel-images. This is an inverse problem requiring data of high accuracy, that is why the system must be well-calibrated.

In the paper we address the problem of evaluation of colour reproduction accuracy in vision systems. The method of such evaluation with use of the newest CIE colour difference formulas (CIE94, CIEDE200) has been proposed. The advantages of the sRGB colour space in colour reproduction on the background of device-dependent RGB colour space are presented. The impact of method of image compression (file format) is also discussed.

The proposed deinterlacing scheme is aimed towards high-quality progressive image output, typically for a videotape-to-film blow-up operation. This operation consists in increasing the resolution of video images, recording these images on 16mm or 35mm film and reducing the typical video artefacts (due to interlaced scanning, electronic noise), eventually adding some film-look features (color correction, motion blur,.). This paper focuses on the . Since video real-time is not relevant for digital blow-up (film recorders are slow), the proposed approach gives more weight to for both of the image. Each field from the interlaced source is extended to a full frame by using an accurate in-painting process, outstretching edges and preserving their intensity. Motion estimation is then performed from these two inpainted images, and the computed forward and backward motion vector fields are finally used for temporal integration. During this process, the inter-field motion is converted to a motion blur appearance, similar to the one physically produced by the prolonged exposure time in film cameras. Finally, the resulting image is upsized and enhanced by a warp-filter, which unlike high pass filters does not augment noise.

This paper presents a family of adaptive color image filters based on robust order-statistics and local contrast entropy control. The proposed filtering class is fully adaptive and free of any optimization procedure. Due to the adaptive nature of the algorithms and excellent balance between noise-removing and detailpreserving characteristics, it is desirable to use this technique as a pre-processing tool in practical color image processing applications such as virtual restoration of artworks.