Fast Motion Estimation Algorithm with Efficient Memory Access for HEVC Hardware Encoders
Pakdaman, Farhad; Gabbouj, Moncef; Hashemi, Mahmoud Reza; Ghanbari, Mohammad (2018-11)
Pakdaman, Farhad
Gabbouj, Moncef
Hashemi, Mahmoud Reza
Ghanbari, Mohammad
IEEE
11 / 2018
2018 7th European Workshop on Visual Information Processing (EUVIP)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-201910234053
https://urn.fi/URN:NBN:fi:tuni-201910234053
Kuvaus
Peer reviewed
Tiivistelmä
The encoding process in the HEVC standard is several times more complex than the previous standards. Since motion estimation is responsible for most of this complexity, the new Test Zone (TZ) search is usually adopted as the fast search algorithm, to alleviate the complexity. However, the TZ search requires a high rate of access to the off-chip memory, which contributes heavily to the total consumed encoding power. In this paper we demonstrate that the process of finding the best starting search point in this algorithm, does not allow effective reduction of memory access in hardware encoders. As a solution, a new fast motion estimation algorithm is proposed which estimates a proper single starting search point, in addition to an adaptively reduced search range, based on available information from the coded neighboring blocks. The experimental results show that this algorithm on average can reduce the required memory access for ME by ~78% and reduce the integer ME time by ~70%, with only 1.1% Bjontegaard Delta (BD) Rate.
Kokoelmat
- TUNICRIS-julkaisut [18544]