21
6

Total variation multiscale estimators for linear inverse problems

Abstract

Even though the statistical theory of linear inverse problems is a well-studied topic, certain relevant cases remain open. Among these is the estimation of functions of bounded variation (BVBV), meaning L1L^1 functions on a dd-dimensional domain whose weak first derivatives are finite Radon measures. The estimation of BVBV functions is relevant in many applications, since it involves minimal smoothness assumptions and gives simplified, interpretable cartoonized reconstructions. In this paper we propose a novel technique for estimating BVBV functions in an inverse problem setting, and provide theoretical guaranties by showing that the proposed estimator is minimax optimal up to logarithms with respect to the LqL^q-risk, for any q[1,)q\in[1,\infty). This is to the best of our knowledge the first convergence result for BVBV functions in inverse problems in dimension d2d\geq 2, and it extends the results by Donoho (Appl. Comput. Harmon. Anal., 2(2):101--126, 1995) in d=1d=1. Furthermore, our analysis unravels a novel regime for large qq in which the minimax rate is slower than n1/(d+2β+2)n^{-1/(d+2\beta+2)}, where β\beta is the degree of ill-posedness: our analysis shows that this slower rate arises from the low smoothness of BVBV functions. The proposed estimator combines variational regularization techniques with the wavelet-vaguelette decomposition of operators.

View on arXiv
Comments on this paper