Sparse Spatial Coding: a novel approach for efficient and accurate object recognition

Abstract

Successful state-of-the-art object recognition techniques from images have been based on powerful techniques, such as sparse representation, in order to replace the also popular vector quantization approach. Recently, sparse coding, which is characterized by representing a signal in a sparse space, has raised the bar on sev-eral object recognition benchmarks. However, one serious drawback of sparse space based methods is that similar local features can be quantized into different visual words. We present in this thesis a new object recognition approach, called Sparse Spa-tial Coding (SSC), which combines a sparse coding dictionary learning and a spatial constraint coding stage. Thus, we minimize the problems of pure sparse represen-tations. Experimental evaluation was done at Caltech 101, Caltech 256, Corel 5000 and Corel 10000, that are datasets specifically designed to object recognition evalu-ation. The obtained results show that, to the best of our knowledge, our approach achieves accuracy beyond the best single feature method previously published on the databases. The method also outperformed, for the same bases, several methods that use multiple feature, and provide equivalent to or slightly lower results than other techniques. Finally, we verify our method generalization, applying the SSC to recognize scene in the Indoor 67 scene dataset, VPC and COLD, displaying perfor-mance comparable to state-of-the-art approaches in the first two bases and superior in COLD dataset.