Texture fidelity benchmark

Abstract

Automatic texture quality evaluation is important but still unsolved difficult problem. While several generative mathematical texture models were developed, their reliable qualitative evaluation is for now possible only using impractical and expensive visual psycho-physics which hampers their further progress. We present the texture fidelity benchmark created to help the validation of texture fidelity criteria being developed. The benchmark is a web based service (http://tfa.utia.cas.cz) designed for performance evaluation, mutual comparison, and ranking of various texture fidelity measures. The benchmark supports rapid verification and development of new fidelity criteria approaches and contains seven color, variable texture quality, series together with their grey-scale counterparts.