Compressive Sensing Based Face Detection without Explicit Image Reconstruction Using Support Vector Machines

Abstract

The novel theory of compressive sensing takes advantage of the sparsity or compressibility of a signal in a specific domain allowing the assessment of its full representation from fewer measurements. In this work we tailored the concept of compressive sensing to assess the intrinsic discriminative capability of this method to distinguish human faces from objects. Afterwards we enrolled through a feature selection study to empirically determine the minimum amount of measurements required to properly detect human faces. This work was concluded with a comparative experiment against the SIFT descriptor. We determined that using only 40 measurements conducted by compressing sensing one is capable of capturing the relevant information that enable one to properly discriminate human faces from objects.