Object recognition via compressive sensing

When the compressive sensing theory is applied to compressive imaging of natural images, we get fewer non-pixel measurements instead of pixel values. We propose how to make the direct use of compressed measurements without time-consuming image recovery for object recognition. For this, we use an orderly-permuted Hadamard matrix as a sensing matrix and extract robust local features from compressed measurements. We show that this challenging work is successful despite that those local features are extracted from limited source of information.