Stochastic-Optimized Low-Latency Rate Control for Image Mezzanine Compression

Image mezzanine compression refers to a class of lightweight compression techniques characterized by ultra-low latency, low complexity, and visually lossless quality. It is widely applied in scenarios such as video transport over digital links, professional video production workflows, and memory buffers. As these scenarios typically involve constant bitrate (CBR) transmission channels, existing image mezzanine compression standards generally incorporate a rate buffer to smooth the bitrate fluctuations, along with corresponding rate control (RC) algorithms to maintain buffer stability. However, this buffer size is typically tightly constrained due to the stringent low-latency requirements. Consequently, current mezzanine compression RC methods rely on heuristic buffer-feedback-based strategies, leaving potential for improvements in rate-distortion (R-D) performance. In this paper, we propose a stochastic optimization-motivated, low-latency RC algorithm for image mezzanine compression. First, we formulate the RC problem as a stochastic optimization framework, explicitly optimizing R-D performance under buffer stability constraints. Second, we propose a Lyapunov optimization-based rate control (LORC) algorithm, which solves this problem with zero lookahead, ensuring low latency. To enhance content and bitrate adaptability, we further develop a λ-domain analysis-based parameter adaptation strategy. The proposed LORC algorithm is implemented in the JPEG XS international mezzanine compression standard. Experimental results demonstrate significant R-D performance gains under the JPEG XS standard-compliant buffer-model setting, with an average BD-PSNR improvement of 1.69 dB and a BD-rate reduction of 10.48%, while introducing only a moderate increase in computational complexity.