Exploring two spaces with one feature: Kernelized multidimensional modeling of visual alphabets