Object Recognition Using Learning-based Compressive Sensing

Abstract

Object recognition plays an important role in many applications, and numerous approaches have been proposed to increase recognition accuracy. However, most of these existing methods use high-dimensional images as an input for recognition algorithms, which imposes a computational burden on the algorithms and limits recognition efficiency. This paper proposes an object recognition approach using learning-based compressive sensing. This method does not require any reconstruction of images and recognizes objects directly using compressed data. Experimental results have demonstrated the effectiveness of this approach and show that a learned sensing matrix has better recognition performance than the traditional random sensing matrix.