Evaluating Focal Stack with Compressive Sensing

Abstract

Compressive sensing (CS) has demonstrated the ability in the field of signal and image processing to reconstruct signals from fewer samples than prescribed by the Shannon-Nyquist algorithm. In this paper, we evaluate the results of the application of a compressive sensing based fusion algorithm to a Focal Stack (FS) set of images of different focal planes to produce a single All In-Focus Image. This model, tests l1 -norm optimization to reconstruct a set of images, called a Focal Stack to reproduce the scene with all focused points. This method can be used with any Epsilon Photography algorithm, such as Lucky Imaging, Multi-Image panorama stitching, or Confocal Stereo. The images are aligned and blocked first for faster processing time and better accuracy. We evaluate our results by calculating the correlation of each block with the corresponding focus plane. We also discuss the shortcomings of this simulation as well as the potential improvements on this algorithm.