Adaptive Modeling of Sky for Video Processing and Coding Applications

Video content analysis for stilland moving images can be used for various applications, such as high-level semantic-driven operations or pixel-level contentdependent image manipulation. Within video content analysis, sky regions of an image form visually important objects, for which interesting applications at both mentioned levels can be envisaged. This paper introduces a new algorithm and model for detecting blue-sky areas, with suitable properties for TV video material. The proposed method is capable of robustly detecting various sky appearances, while addressing the requirements of the target applications. A special feature of our proposal is that we use adaptive color 1 and position models for computing a pixel-accurate sky-probability measure. The experimental simulations show that our proposal considerably improves the correct detection/rejection of sky regions, and yields better spatial and temporal consistency of the detection results with respect to the currently available systems.