Adaptive Pixel Resizing for Multiscale Recognition and Reconstruction
Résumé
this paper present an adaptive pixel resizing method based on a parameter space approach. The pixel resizing is designed for both multiscale recognition and reconstruction purposes. The general idea is valid in any dimension. In this paper we present an illustration of our method in 2D. Pixels are resized according to the local curvature of the curve to control the local error margin of the reconstructed Euclidean object. An efficient 2D algorithm is proposed.