Power Consumption of Linear and Non-linear Filters on Multi-core Architecture by Using AVX and OpenMP

The general concept of filtering is to modify or enhance the image data by removing noises for smoothing, sharpening, and edge enhancement. There are mainly two types of filtering methods in image processing, namely linear and non-linear filters. This paper proposes the energy consumption of linear and non-linear filters on multi-core architecture by using AVX and OpenMP. Many researchers have contributed to the computation time of well-known filters such as the mean, median, Gaussian, Laplacian, IRR filters, and so on. Although these filters have been experimented related with to execution time, the measurement of energy consumption of linear and non-linear filters has still remained a gap for image processing. For this reason, this paper focuses on the experiments for the energy consumption of a linear filter (Gaussian filter) and a non-linear filter (median filter). Moreover, the parallel implementation of these filters and the serial computation also have been tested on CPU and GPU in the case of examining the measuring the amount of energy consumption of them. According to the experimental results of the matrix kernel becomes 4069 ×4096, in average, a linear filter is more consumed than a non-linear filter in the case of energy consumption of about 5.1 % and 3.9% in Watt and more efficient performance of about 2.5% and 1.12% in GFLOPS when testing on CPU and GPU respectively, when utilizing AVX and OpenMP in the parallel versions of these filters.