Fast computation of polar harmonic transforms

Polar harmonic transforms (PHTs) are orthogonal rotation invariant transforms that provide many numerically stable features. The kernel functions of PHTs consist of sinusoidal functions that are inherently computation intensive. We develop a fast approach for their computation using recursion and 8-way symmetry/anti-symmetry property of the kernel functions. The clustering of pixels at eight radially symmetrical locations enhances the speed of computation. Experimental results show that the proposed method is faster by a factor lying between three and four compared to the existing fast method.