Image approximation by minimization of a geometric distance applied to a 3D finite elements based model

We propose a method for image coding based on an adaptive non-uniform sampling scheme. The aim is to optimize a 3D mesh-based model such as those involved in finite elements method. To this end, a geometrical distance has been defined, measuring the error between the model and the original image, and a gradient algorithm has been derived to obtain an accurate model deformation. The corresponding 2D rendering lends to an efficient image representation: salient features such as edges are preserved, offering a good subjective quality.