Image processing is a class of procedures very helpful in several research fields. In a general scheme, a starting image generates a output image, or some image features, whose values are composed by using different methods. In particular, among image processing procedures, image restoration represents a current challenge to address. In this context the noise removal plays a central role. Here, we consider the specific problem of stripe noise removal. To this aim, in this paper we propose a novel Gaussian-based method that works in the frequency domain. Due to the large computational cost when using, in general, Gaussian related methods, a suitable parallel algorithm is presented. The parallel implementation is based on a specific strategy which relies the newest powerful of graphic accelerator such as NVIDIA GPUs, by combining CUDA kernels and OpenACC’s routines. The proposed algorithm exhibits good performance in term of quality and execution times. Tests and experiments show the quality of the restored images and the achieved performance.
A Novel GPU Implementation for Image Stripe Noise Removal
De Luca P.;Galletti A.;Marcellino L.
2022-01-01
Abstract
Image processing is a class of procedures very helpful in several research fields. In a general scheme, a starting image generates a output image, or some image features, whose values are composed by using different methods. In particular, among image processing procedures, image restoration represents a current challenge to address. In this context the noise removal plays a central role. Here, we consider the specific problem of stripe noise removal. To this aim, in this paper we propose a novel Gaussian-based method that works in the frequency domain. Due to the large computational cost when using, in general, Gaussian related methods, a suitable parallel algorithm is presented. The parallel implementation is based on a specific strategy which relies the newest powerful of graphic accelerator such as NVIDIA GPUs, by combining CUDA kernels and OpenACC’s routines. The proposed algorithm exhibits good performance in term of quality and execution times. Tests and experiments show the quality of the restored images and the achieved performance.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.