Performance and Microarchitectural Analysis for Image Quality Assessment

Abstract

This thesis presents performance analysis for five matured Image Quality Assessment algorithms: VSNR, MAD, MSSIM, BLIINDS, and VIF, using the VTune ... from Intel. The main performance parameter considered is execution time. First, we conduct Hotspot Analysis to find the most time consuming sections for the five algorithms. Second, we perform Microarchitecural Analysis to analyze the behavior of the algorithms for Intel's Sandy Bridge microarchitecture to find architectural bottlenecks. Existing research for improving the performance of IQA algorithms is based on advanced signal processing techniques. Our research focuses on the interaction of IQA algorithms with the underlying hardware and architectural resources. We propose techniques to improve performance using coding techniques that exploit the hardware resources and consequently improve the execution time and computational performance. Along with software tuning methods, we also propose a generic custom IQA hardware engine based on the microarchitectural analysis and the behavior of these five IQA algorithms with the underlying microarchitectural resources.