Hardware Accelerator of Discrete Cosine Transformation for JPEG Algorithms

U ovom je radu obrađena izvedba diskretne kosinusne transformacije (engl. Discrete Cosine Transform, u nastavku: DCT) na razvojnoj pločici ZedBoard Zynq 7000 SoC(u nastavku: Zynq), s posebnim naglaskom na važnost DCT-a za JPEG normu kompresije digitalnih slika. JPEG algoritam koristi činjenice da je ljudsko oko osjetljivije na razlike u svjetlini nego u tonu boje, te da je ljudsko oko manje osjetljivo na visokofrekventne komponente slike. DCT omogućava njihovo komprimiranje ili čak izbacivanje iz konačnog rezultata. Za sklopovsku je izvedbu DCT-a korišten Zynq u sinergiji s njegovom programskom podrškom – paketom programa Vivado. Dio tog paketa, Vivado HLS, omogućio je stvaranje intelektualnog vlasništva koje vrši DCT uz određen stupanj optimizacije; Vivado je omogućio spajanje s pločicom, dok je Vivado SDK korišten za sâmo pokretanje algoritma na Zynqu. Zbog tehničkih poteškoća na pločici je izveden samo DCT, a ne i cijeli JPEG algoritam, ali iz postignutog je očito da je moguće dobiti ubrzanja zahtjevnih izračuna korištenjem sklopovskog ubrzanja.This thesis deals with the implementation of discrete cosine transform (DCT) on the ZedBoard Zynq 7000 SoC development board, with emphasis on DCT's importance for the JPEG image compression standard. The JPEG standard makes use of the human eye's inherent greater sensitivity to luminous differences in colours than to their chrominant counterparts, along with the eye's weaker sensitivity to high-frequency image regions. DCT makes it possible to greatly compress these components, or even eliminate them completely. The implementation of DCT on the hardware was made possible by using a software package called Vivado. Vivado HLS was used to create the intellectual property (IP) which handled DCT with a certain degree of optimisation; Vivado was used to connect the IP to the board itself, and Vivado SDK was used to actually run the program on the board. Due to technical difficulties, only DCT was implemented, not the whole JPEG algorithm, but even this degree of success suggests it is possible to greatly improve the speed of time-consuming calculations using hardware acceleration