04 - Apport de l'espace Teinte-Saturation-Luminance pour la segmentation spatiale et temporelle
Sébastien Lefèvre, Nicole Vincent · HAL (Le Centre pour la Communication Scientifique Directe) · 2006
In this article, we focus on the Hue Saturation Luminance colour space, known for its appropriate representation of the human vision, and we illustrate its interest in the segmentation of image data, both in the spatial and temporal dimensions. The HSL space, with its particular properties (one angular component, the hue, and two scalar components, the saturation and the luminance), requires some original computation models. So our contribution is triple : first we will deal with colour representation in the HSL space and with specific computation methods which have to be involved. Then we will show the relevance of this colour space for automatic segmentation of image data in the two domains which are space and time. Spatial segmentation is considered under the problem of background and foreground separation for which we propose a multiresolution approach which requires a single image. Temporal segmentation corresponds to shot change detection in an image sequence, and the method we are proposing is based on the use of context-independent distance measures. Common properties of our both methods (spatial and temporal segmentations) are efficiency (processing time compatible with video framerate) and robustness (in particular against illumination changes). We illustrate these two approaches with results obtained in the domain of sport video sequences analysis and we compare in this context the use of HSL and RGB colour spaces.