We present the Optimizing Control Variate (OCV) estimator, a new estimator for Monte Carlo rendering. Based upon a deterministic sampling framework, OCV allows multiple importance sampling functions to be combined in one algorithm. Its optimizing nature addresses a major problem with control variate estimators for rendering: users supply a generic correlated function which is optimized for each estimate, rather than a single highly tuned one that must work well everywhere. We demonstrate OCV with both direct lighting and irradiance-caching examples, showing improvements in image error of over 35% in some cases, for little extra computation time.

Shaohua Fan, Stephen Chenney, Bo Hu, Kam-Wah Tsui, Yu-chi Lai.* Optimizing Control Variate Estimators for Rendering*, Computer Graphics Forum, Vol. 25, No. 3, pp. 351-358, 2006.