Coding scene contents using the image background

We propose a scene coding approach that allows us to maintain the spatial relations among the individual objects and between these objects and the background by a suitable representation of their shapes. To this aim we apply the recently introduced Boundary Integral - Resonant Mode Expansion (BI-RME) approach for shape analysis, whose aim is to compute the vibration modes of an elastic sheet of fixed boundary to represent an arbitrarily shaped object. We demonstrate that the BI-RME algorithm is suitable to give contemporary and efficiently the shape representation for all the objects present in the original image. Moreover, it preserves the spatial relations by the analysis of the scene background, i.e. the part of an image left when all the meaningful objects have been extracted.