Combined kernel for fast GPU computation of Zernike moments

Zernike moments, as a representative orthogonal moment, have been widely applied in the fields of image processing and pattern recognition. The calculations are time-consuming due to the complexity of definition. Based on the GPU octant symmetry algorithm in our previous work, this paper presents a novel algorithm to increase the resource utilization by the combined kernel. Also, it optimizes radial polynomials of Zernike moments to reduce amount of calculations. The experimental results demonstrated that the proposed algorithm achieved overall computational performance improvement for any sized images. Moreover, there is no compromise in terms of precision compared to the typical accurate algorithm.