Fast Early Complexity Estimation of Test Zone Search Algorithm in VVenC

Abstract

The Versatile Video Coding (VVC) standard provides significant compression gains, but practical encoder implementations introduce a substantial computational burden. In particular, the Motion Estimation (ME) process in encoders such as VVenC relies on highly complex algorithms, including the Test Zone Search (TZS). Since existing complexity prediction metrics often require costly preprocessing, this work proposes $C_{M V P}$, a low-cost prediction method based on the early stages of the TZS algorithm in VVenC. The metric exploits the spatial variance and dispersion of motion vectors to anticipate subsequent computational costs. Results show that $C_{M V P}$ achieves a moderate to strong Pearson Correlation Coefficient (PCC) with TZS memory accesses and execution time. Moreover, it outperforms traditional spatial metrics $(E$ and SI) while maintaining a consistently low computational overhead across all evaluated resolutions. These findings indicate that early complexity estimation through $C_{M V P}$ provides a promising foundation for strategies that balance coding efficiency and visual quality across diverse scenarios.