Sparse Representation-based Object Recognition - Sparse Representations and Compressive Sensing for Imaging and Vision - page 82

The performance of kerSMBR and kerSLR are better than their linear counterparts,
however kerSVM does not show much improvement. Composite kernels present an
interesting case. Here, compSLR shows better performance than compSMBR on all
the modalities. Composite kernel by combining homogenous modalities into one,
reduces effective number of modalities, hence the size of

matrix is reduced. This
decreases the flexiblity in exploiting different modality information via

Γ

Γ . Hence,

the performance of compSMBR is not optimal.