Compressive image feature extraction by means of folding

Abstract

We explore the utility of a dimensionality reducing process we term folding for the purposes of image feature extraction. We seek to discover whether image features are preserved under this process and how to efficiently extract them. The application is in size weight and power constrained imaging scenarios where an efficient implementation of this dimensionality reduction can save power and computation costs. The specific features we explore are image corners, rotation, and translation. We present algorithms for recovering these features from folded representations of images followed by simulation results showing the performance of the algorithms when operating on real images.