Parallel Computation for Discrete Orthogonal Moments of Images Using Graphic Processing Unit

摘要

A novel method is proposed for fast computing discrete orthogonal moments of large scale digital images using CUDA (Compute Unified Device Architecture) on GPU (Graphic Processing Unit). After original input image loading and mapping by partition model, parallelism was implemented by dividing onto GPU. Experimental results show that the proposed method outperforms the existing software implementation approaches, such as direct method, recursive algorithm based on CPU and so on, especially for larger images and higher order moments which can be performed in real-time.