Many important imaging problems have complex structures that challenge the design of their algorithmic solutions. Contemporary methods for addressing such intricate problems are often formed as conceptually-simple iterative procedures that rely on well-engineered black-box modules, usually in the form of existing denoising techniques.
This talk overviews an alternative optimization framework to address intricate imaging problems based on prominent compression standards as the fundamental modules iteratively applied. The proposed algorithms utilize the alternating direction method of multipliers (ADMM) optimization technique to address seemingly impractical tasks, involving aspects of acquisition, compression and rendering of images. Specifically, we demonstrate how the state-of-the-art HEVC compression standard can be adjusted to several layouts of acquisition and rendering systems, such that the end-to-end reconstruction errors of the systems are optimized with respect to the compression bit-costs. The experiments established our method as the best approach for optimizing the system performance at high bit-rates from the compression standpoint.
Moreover, we relate the proposed approach also to image restoration using complexity regularization, where the likelihood of candidate solutions is evaluated based on their compression bit-costs. Our proposed methods restore signals by leveraging state-of-the-art models designed for compression. The presented experiments show good results for image deblurring and inpainting using the JPEG2000 and HEVC compression standards.
This is joint work with Prof. Alfred Bruckstein and Prof. Michael Elad.