Recovery-Oriented Optical System Design Based on Fisher Information

Abstract

Traditional optical system design for imaging applications has long been guided by image quality metrics such as geometrical RMS spot size, wavefront error, and modulation transfer function (MTF). Although these metrics effectively characterize direct imaging performance, they do not necessarily indicate optimal behavior when image restoration is subsequently applied. In modern imaging pipelines, however, optical systems are increasingly combined with physics-based or learning-based restoration methods, making the information transfer properties of the optical system critically important. In this work, we propose a recovery-oriented optical design approach that explicitly links optical system optimization with image restoration performance through an estimation-theoretic framework based on Fisher information. A new design criterion is adopted to ensure that the information encoded by the optical system can be reliably recovered during image restoration. To validate the proposed approach, two dual separated aspheric lenses are designed under identical configurations using conventional RMS-based optimization and the proposed recovery-oriented optimization, respectively. Although the RMS-optimized lens exhibits superior direct imaging quality, it underperforms after image restoration compared to the recovery-oriented design. These results demonstrate that recovery-oriented optical design provides a physically interpretable and effective extension of traditional optical design criteria for imaging systems operating within modern computational imaging pipelines.