SIMD-based low bit-depth motion estimation with application to HEVC


DUVAR R., KÜÇÜKMANİSA A., AKBULUT O., Celebi A., URHAN O.

SIGNAL IMAGE AND VIDEO PROCESSING, cilt.17, sa.4, ss.1449-1456, 2023 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 17 Sayı: 4
  • Basım Tarihi: 2023
  • Doi Numarası: 10.1007/s11760-022-02353-6
  • Dergi Adı: SIGNAL IMAGE AND VIDEO PROCESSING
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, zbMATH
  • Sayfa Sayıları: ss.1449-1456
  • Anahtar Kelimeler: HEVC, Motion estimation, SIMD, EFFICIENT HARDWARE ARCHITECTURE, SEARCH ALGORITHM, 2-BIT TRANSFORM
  • Kocaeli Üniversitesi Adresli: Evet

Özet

Motion estimation (ME) is a crucial stage in video encoding process since it is the main contributor to high compression ratios. However, it is a very time-consuming operation when a full-search-based block-motion estimation approach is adopted in the encoding process. There are many low bit-depth representation ME approaches in the literature which aim to speed-up this processing without scarifying encoding quality because of their possible efficient hardware implementations. In this paper, we present a single instruction multiple data (SIMD)-based methodology specifically designed for low bit-depth ME approaches and show that this approach can be efficiently implemented in software. The proposed method provides encoding time-savings, up to 68%, according to the full search-based motion estimation approach and has less encoding time, up to 15%, with respect to the fast sparse search-based motion estimation approach. Our experiments on state-of-the-art video encoding standard HEVC clearly show that low bit-depth approaches can have potential in software implementations as well.