Regularization tools and models for image and signal reconstruction

摘要

The present paper proposes a synthetic overview of regulatization techniques for the reconstruction of piecewise regular signals and images. The stress is put on STkhonov penalized approch and on subsequent non-quadratic and halfquadratic generalizations. On one had, a link is made between the detection-estimation formulation and the non-convex penalization approach. On the other hand, it is highighted that convex penalizing functions provide a good edge-preserving compromise between quadratic regularization and the numerically burdensome detection-estimation approach.