On The Performance of Human Visual System Based Image Quality Assessment Metric Using Wavelet Domain A. Ninassi a,b , O. Le Meur a , P. Le Callet b and D. Barba b a Thomson Corporate Research, 1 Avenue Belle Fontaine 35511 Cesson-Sevigne, France; b IRCCyN UMR 6597 CNRS, Ecole Polytechnique de l’Universite de Nantes rue Christian Pauc, La Chantrerie 44306 Nantes, France ABSTRACT Most of the efficient objective image or video quality metrics are based on properties and models of the Human Visual System (HVS). This paper is dealing with two major drawbacks related to HVS properties used in such metrics applied in the DWT domain : subband decomposition and masking effect. The multi-channel behavior of the HVS can be emulated applying a perceptual subband decomposition. Ideally, this can be performed in the Fourier domain but it requires too much computation cost for many applications. Spatial transform such as DWT is a good alternative to reduce computation effort but the correspondence between the perceptual subbands and the usual wavelet ones is not straightforward. Advantages and limitations of the DWT are discussed, and compared with models based on a DFT. Visual masking is a sensitive issue. Several models exist in literature. Simplest models can only predict visibility threshold for very simple cue while for natural images one should consider more complex approaches such as entropy masking. The main issue relies on finding a revealing measure of the surround influences and an adaptation: should we use the spatial activity, the entropy, the type of texture, etc.? In this paper, different visual masking models using DWT are discussed and compared. Keywords: Quality Assessment, Human Visual System, DWT, DFT, Contrast Masking, Entropy Masking 1. INTRODUCTION The aim of an objective image quality assessment is to create an automatic algorithm that evaluates the picture or video quality as a human observer would do. Image quality assessment has been extensively studied during this past few decades and many different objective criteria have been built. The quality metrics based on models of the Human Visual System (HVS) are an important part of the different approaches in image quality assessment. HVS models may be classified into mono-channel or multi-channel models. This work focus on the latter. In order to simulate the multi-channel behavior of the HVS and to well qualify the visual masking effects, this kind of quality metrics rests on a perceptual subband decomposition. This decomposition is often realized in the frequency domain like Fourier domain. The use of Fourier domain leads to good performances, but with high computational complexity. One solution to deal with this would be to use a spatial transform like a wavelet transform. Nevertheless, the correspondence between the visual system and the wavelet domain is known to be only approximate 1 . 2 This issue results in non-optimal visual performance, especially in the setting up of the visual masking. But even if the visual masking effects are non-optimal, what is the decrease in performance in terms of quality assessment? In this paper, the performances loss between an image quality metric using Fourier domain and an image quality metric using wavelet domain are evaluated. An efficient image quality metric based on a multi-channel model of the HVS using wavelet domain is described. This metric provides quality scores well correlated with those given by human observers. The HVS model of the low-level perception used in this metric includes subband decomposition, spatial frequency sensitivity, contrast masking and entropy masking. The subband decomposition of this multi-channel approach is based on a spatial frequency dependent wavelet transform. Advantages and limitations of the wavelet transform as a part of a HVS model are discussed, and compared with a HVS model based on a Fourier transform. The spatial frequency sensitivity of the HVS is simulated by a wavelet contrast sensitivity function (CSF) derived from Daly’s CSF. 3 Masking effects include both contrast masking and entropy masking. Entropy masking allows to consider the modification of the visibility