VSI metric for objective image quality assessment

Kako su subjektivne metode skup i dugotrajan proces za ocjenjivanje kvalitete slike koji se ne može provesti u stvarnom vremenu, koriste se objektivne metode od kojih su najpoznatije PSNR (engl. Peak Signal to Noise Ratio) i SSIM (engl. Structural similarity index). Posljednjih nekoliko godina provedena su mnoga istraživanja kako bi se došlo do zaključka koji dijelovi slika privlače najviše pažnju ljudskog vizualnog sustava. U ovom je radu predstavljena objektna metrika imena VSI (engl. Visual saliency-based index), koja koristi informacije tih istraživanja. Provedeni su eksperimenti ocjene kvalitete dijela slika iz VCL@FER baze slika, te su ti rezultati uspoređivani s rezultatima PSNR i SSIM metrika. Iako je za odabrane slike VSI pokazala slabije performanse od SSIM metrike, za puno veće skupove podataka VSI se ipak pokazala kao puno bolja metrika u predviđanju stvarne kvalitete slike.Subjective methods are expensive and time-consuming process for evaluation image quality and cannot be implemented in real time systems. That’s why objective methods most notably Peak Signal to Noise Ratio (PSNR) and Structural similarity index (SSIM) are being used. Over the last few years many studies have been conducted in order to obtain the conclusion which image sections attract the most attention of the human visual system. This paper presents the objective metric name Visual sailency-based index (VSI), which uses information obtained from these studies. Image quality evaluation experiments were performed on set of images from VCL @ FER database and these results were compared to the results of PSNR and SSIM metrics. Although for selected images VSI showed weaker performance than SSIM, VSI proved to be a much better metric to predict the actual image quality for much larger data set