1: \begin{abstract}

2: A simple greedy refinement procedure for the generation of data-adapted

3: triangulations is proposed and studied. Given a function $f$

4: of two variables, the algorithm produces a hierarchy of triangulations

5: $(\cD_j)_{j\geq 0}$ and piecewise polynomial approximations 

6: of $f$ on these triangulations. The refinement procedure consists in bisecting

7: a triangle $T$ in a direction which is chosen so as to minimize the local approximation error

8: in some prescribed norm between $f$ and its piecewise polynomial approximation

9: after $T$ is bisected. The hierarchical structure allows us to derive

10: various approximation tools such as multiresolution analysis, wavelet bases,

11: adaptive triangulations based either on greedy or optimal

12: CART trees, as well as a simple encoding of the corresponding triangulations.

13: We give a general proof of convergence in the $L^p$ norm

14: of all these approximations.

15: Numerical tests performed in the case of piecewise linear

16: approximation of functions with analytic expressions or of numerical images 

17: illustrate the fact that the refinement procedure generates triangles with an optimal aspect ratio (which

18: is dictated by the local Hessian of $f$ in case of $C^2$ functions).

19: \end{abstract}

20: