Overview of algorithms for incresing of video temporal resolution

Radi očuvanja mrežnih resursa i omogućavanja dovoljno brzog prijenosa videosignala u uvjetima malih brzina prijenosa potrebno je smanjiti količinu podataka za prijenos. Smanjenje količine podataka za prijenos izvršava se kodiranjem videosignala s ciljem smanjenja vremenske rezolucije. Nakon prijenosa mrežom istim video sadržajima potrebno je povećati vremensku rezoluciju kako bi bili prikladni za prikaz na uređajima koji podržavaju videosadržaj visoke prostorne i vremenske rezolucije. U radu je ispitana kvaliteta interpoliranih slika dobivenih korištenjem nekih od postojećih algoritama za povećanje vremenske rezolucije videosignala. Za analizu su odabrana dvije algoritma koja koriste isključivo prostorne značajke videosignala (FA i FR) te dva algoritma koji koriste prostorne i vremenske značajke videosignala (AST-LS i AST-FS). U okviru usporedbe algoritama koji koriste samo prostorne značajke videosignala uočena je prednost u kvaliteti interpoliranih slika sa FA algoritmom uz približno jednaku računalnu zahtjevnost i vrijeme izvedbe. FA algoritam je pokazao najveću razinu kvalitete interpolirane slike za videosadržaje male vremenske aktivnosti dok za sadržaje velike vremenske aktivnosti prednjače algoritmi koji koriste vremenske i prostorne značajke videosignala. Usporedbom AST-LS i AST-FS algoritama kao bolje rješenje pokazao se AST-LS zbog manje računalne kompleksnosti uz gotovo jednake rezultate kvalitete interpoliranih slika. Za objektivnu ocjenu interpoliranih slika korištene su PSNR i SSIM metrike. For the purpose of efficient usage of network resources and enabling fast transmission it is needed to reduce amount of video signal data. Amount of the data for transmission is reduced by reducing temporal resolution in the coding process. After transmission over the network, it is needed to perform algorithms for frame rate-up conversion to make the video signal suitable for reproduction on devices supporting videos with high spatial and temporal activity. In this thesis, quality of interpolated images was measured. Interpolated images were made with some of already proposed algorithms for frame rate up-conversion. Two of the analyzed algorithms (FR and FA) use only spatial features of video signal and other two algorithms use spatial and temporal features of video signal (AST-FS and AST-LS). The FA algorithm shows much better quality of interpolated images compared to the FR with similar computational complexity and execution time. In case of very slow video sequences, the FA algorithm shows the best results of interpolated image quality but in case of very fast video sequences, algorithms that use spatial and temporal features of video signal are leading. The AST-LS, in comparison with the AST-FS, was revealed as a better decision because of similar quality of interpolated images but less computationally complex. PSNR and SSIM metrics were used for objective interpolated image evaluation.