The Improvement of 1D Gaussian Blur Filter using AVX and OpenMP
Nwe Zin Oo · 2022 22nd International Conference on Control, Automation and Systems (ICCAS) · 2022
A Gaussian blur filter is applied to produce the new pixel value of an image using a Gaussian function. It is a low-pass filter and modifies the input image pixels by convolution the matrix with a Gaussian function which becomes computationally intensive work for image processing. In this paper, the examination of performance, energy/power consumption between the original 1D, 2D, and modified 1D convolution used in the Gaussian blur filter is proposed by using Advanced Vector Extension (AVX) intrinsics, loop unrolling, and OpenMP on multi-core architecture. The main contribution aims to parallelize the modified 1D convolution on parallel processing cores to enhance the performance and reduce the processing time by focusing on the Gaussian filter function for contrast the original 1D and 2D convolution. The improved performance gains approximately 1.3 times, and 6.9% in energy and 3.2% in power consumption is reduced respectively when testing on Intel core i7 multi-core while keeping up the energy awareness of different filtering kernels.