Adaptive Parameter Computing on Structure-Texture Image Decomposition

Image decomposition is an important inverse problem in image processing. PDE-based decomposition is employing the total variation energy to split an original image into the structural part and textural part. The result greatly relies on the weight parameter which balances these two parts, big or small may produce too much noise or over-smoothness. In this paper, we discuss an adaptive parameter computing method to choose the weight for the image with some periodicity and oscillatory textures, use a TV-Hilbert regularization model and Chambollepsilas projection algorithm to produce the decomposition, and present some numerical results.