Object classification with multi-scale autoconvolution

This paper assesses the recently proposed affine invariant image transform called a multi-scale autoconvolution (MSA) in some practical object classification problems. A classification framework based on the MSA and support vector machines is introduced. As shown by the comparison with another affine invariant technique, it appears that this new technique provides a good basis for problems where the disturbances in classified objects can be approximated with spatial affine transformation. The paper also introduces a new property clarifying the parameter selection in the multi-scale autoconvolution.