PISTA: Parallel Iterative Soft Thresholding Algorithm for Sparse Image Recovery

Attilio Fiandrotti, Sophie M. Fosson, Chiara Ravazzi, and Enrico Magli

30th Picture Coding Symposium, December 8-11, San Jose, California, USA

Abstract We present PISTA, a GPU-accelerated Iterative Soft Thresholding (IST) algorithm for sparse image recovery in Compressive Sensing applications. As the time required to recover an image increases with the number of pixels, GPU-acceleration enables to recover even large images in reasonable time. With respect to equivalent methods, IST-like algorithms have lower computational complexity per-iteration and lower memory requirements, plus the operations are inherently suitable for parallelization. Our experiments show that our algorithm enables a significant reduction in the time required to recover an image even over a highly-optimized CPU-only reference.

Additional material Click on an item to open a preview, then on  (top-left) to download it. pista

Presentation