Analysis of the Difference of Gaussians Model in Image Difference Metrics

Sebastien Akli Ajagamelle1,3, Marius Pedersen2,3, Gabriele Simone3 1Grenoble Institute of Technology, 2Océ Print Logic Technologies, France, 3Gjøvik University College, Norway Download paper Play (17min) Download: FLV | MOV | OGG | MP3

The goal of this work is to present and review two new image difference metrics, named SDOG-CIELAB and SDOG-DEE. These metrics are along the same lines as the standard S-CIELAB metric (Zhang and Wandell, 1997), modified to include a pyramidal subsampling, the Difference of Gaussians receptivefield model (DOG) (Tadmor and CGIV 2010 and MCS’10 Final Program and Proceedings xxvii Tolhurst, 2000), and the ΔEE color-difference formula (Oleari et al., 2009). The DOG model and the ΔEE formula have been shown to improve respectively contrast measures and image quality metrics (Simone et al., 2009). Extensive testing using 29 stateof- the-art metrics and six image databases has been performed. Although this new approach is promising, we only find weak evidence of effectiveness. Analysis of the results indicates that the metrics show fairly good correlations over particular test images, yet they do not outperform the most common objective quality measures.

You may also like:

1. Noise Analysis of a Multispectral Image Acquisition System
2. Comparison of Colour Difference Methods for Natural Images
3. Checking Recent Colour-Difference Formulas with a Dataset of Metallic Samples and Just Noticeable Colour-Difference Assessments
4. Evaluating Color Difference Formulae by Riemannian Metric
5. Spectral Image Prediction of Color Halftone Prints Based on Neugebauer Modified Spectral Reflection Image Model