Detail aplikovaného výsledku

Hardware Accelerator for Evolutionary Image Filters Design

VAŠÍČEK, Z.; SEKANINA, L.

Originální název

Hardware Accelerator for Evolutionary Image Filters Design

Anglický název

Hardware Accelerator for Evolutionary Image Filters Design

Druh

Funkční vzorek

Abstrakt

The EHWFILTER accelerator was developed in order to accelerate image filter evolution in hardware. It can automatically create a filter (e.g., to suppress shot noise) when noised image and uncorrupted original image are provided. The accelerator is implemented on the COMBO6X card equipped with Virtex II Pro 2VP50ff1517 FPGA. It consists of genetic unit, fitness unit and the so-called virtual reconfigurable circuit (VRC) which is utilized to evaluate candidate filters. Every image filter is considered as a digital circuit of nine 8-bit inputs and a single 8-bit output, which processes grayscale (8-bits/pixel) images. Training images are stored in external SRAM memories. The search algorithm is based on Cartesian Genetic Programming operating over 4x8 programmable elements, population size of 8 individuals, and 1 mutation/chromosome. This setting is default but can be changed. The accelerator is connected with PC using PCI bus. The system requires approx. 10 sec to produce a filter which is 44 times faster than the same algorithm running at the Celeron 2.4 GHz. 

Abstrakt aglicky

The EHWFILTER accelerator was developed in order to accelerate image filter evolution in hardware. It can automatically create a filter (e.g., to suppress shot noise) when noised image and uncorrupted original image are provided. The accelerator is implemented on the COMBO6X card equipped with Virtex II Pro 2VP50ff1517 FPGA. It consists of genetic unit, fitness unit and the so-called virtual reconfigurable circuit (VRC) which is utilized to evaluate candidate filters. Every image filter is considered as a digital circuit of nine 8-bit inputs and a single 8-bit output, which processes grayscale (8-bits/pixel) images. Training images are stored in external SRAM memories. The search algorithm is based on Cartesian Genetic Programming operating over 4x8 programmable elements, population size of 8 individuals, and 1 mutation/chromosome. This setting is default but can be changed. The accelerator is connected with PC using PCI bus. The system requires approx. 10 sec to produce a filter which is 44 times faster than the same algorithm running at the Celeron 2.4 GHz. 

Klíčová slova

accelerator, FPGA, Combo6X, evolutionary algorithm, image filter

Klíčová slova anglicky

accelerator, FPGA, Combo6X, evolutionary algorithm, image filter

Umístění

Ústav počítačových systémů, Fakulta informačních technologií VUT v Brně, Božetěchova 2, 612 66 Brno, http://www.fit.vutbr.cz/units/UPSY/

Možnosti využití

výsledek využívá pouze poskytovatel

Licenční poplatek

Využití výsledku jiným subjektem je možné bez nabytí licence (výsledek není licencován)

www

Dokumenty