Abstract
Image and video processing applications require various 2D filters, many of them separable filters. This separable computation is an advantage since it implies less arithmetic computational complexity. But a direct hardware implementation of separable filter requires huge memory to store the results of the first 1D filter, which are not available in FPGAs or hard to implement in ASICs. Thus external memory access is required, which results in reduced speed due to the memory accessing overheads. This paper introduces an alternative approach with overlapped block processing, only requiring a memory size that is approximately equal to the image block size, thus no external memory access is required. As an example we implement the gradient calculation of the Sobel operator in a FPGA.
Original language | English |
---|---|
Title of host publication | Electronics, Communications and Networks IV |
Subtitle of host publication | Proceedings of the 4th International Conference on Electronics, Communications and Networks, CECNet2014 |
Editors | Amir Hussain, Mirjana Ivanovic |
Place of Publication | Boca Raton, FL |
Publisher | CRC Press/Balkema |
Pages | 1355-1358 |
Number of pages | 4 |
ISBN (Electronic) | 9781315682105 |
ISBN (Print) | 9781138028302 |
Publication status | Published - 2015 |
Event | 4th International Conference on Electronics, Communications and Networks, CECNet2014 - Beijing, China Duration: 12 Dec 2014 → 15 Dec 2014 |
Other
Other | 4th International Conference on Electronics, Communications and Networks, CECNet2014 |
---|---|
Country/Territory | China |
City | Beijing |
Period | 12/12/14 → 15/12/14 |