FPGA implementation of machine learning based image quality assessment

Abstract

This paper presents the construction and implementation process on an FPGA platform of an objective perceived image quality, using an objective image quality assessment (IQA) method. This objective IQA uses machine learning (ML) methods to construct the models upon the features extracted from different concepts: the natural scene statistic (NSS) in spatial domain, the gradient magnitude (GM), the Laplacian of Gaussian (LoG), as well as the spectral and spatial entropies. The training phase to estimate the image quality is performed by a learning which uses two training phases to predict the objective image quality; the first to train the intermediary metrics using the classes of independent features, and the second to evaluate the image quality using the intermediary metrics. The Implementation phase on an Field Programmable Gate Array (FPGA) platform is tested on Xilinx Virtex 7 (VC707) FPGA board, and implemented using C/C++ code on Xilinx Vivado HLS.