DocumentCode :
3379849
Title :
3-neighborhood motion estimation in CNN silicon architectures
Author :
Koskinen, Lauri ; Paasio, Ari ; Halonen, Kari
Author_Institution :
Electron. Circuit Design Lab., Helsinki Univ. of Technol., Finland
Volume :
5
fYear :
2004
fDate :
23-26 May 2004
Abstract :
A 3-neighborhood connection for CNN silicon architectures is presented. The 3-neighborhood connection is ideal for computing the matching criterion, such as MSE or SAD, needed in the motion estimation of current video coding standards such as MPEG-4 or H.264. The 3-neighborhood can achieved eight additional connections. An integrated chip with two 9 × 9 CNN test arrays capable of MSE and SAD calculation was implemented in 0.18 CMOS. Simulations in MPEG-4 show comparable performance to fully digital solutions. In worst case SAD accuracy conditions there is only under a 0.2 dB drop in PSNR and under a 1.3% increase in coded bits. With the MSE the figures are 0.1 dB and 0.71%.
Keywords :
CMOS integrated circuits; cellular neural nets; integrated circuit interconnections; mean square error methods; motion estimation; silicon; 3-neighborhood connection; 3-neighborhood motion estimation; 9 × 9 CNN test arrays; CMOS implementation; CNN silicon architectures; H.264; MPEG-4; MSE calculation; PSNR; SAD calculation; Si; coded bits; digital solution; integrated chip; matching criterion computing; video coding standards; Cellular networks; Cellular neural networks; Computer architecture; Computer networks; Concurrent computing; MPEG 4 Standard; Motion estimation; Sensor arrays; Silicon; Video coding;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Circuits and Systems, 2004. ISCAS '04. Proceedings of the 2004 International Symposium on
Print_ISBN :
0-7803-8251-X
Type :
conf
DOI :
10.1109/ISCAS.2004.1329906
Filename :
1329906
Link To Document :
بازگشت