Electrically tunable liquid-crystal (LC) lenses enable compact and energy-efficient variable-focus imaging systems. However, image degradation such as defocus blur and contrast reduction typically occurs during lens operation. In this study, a deep-learning-based restoration approach using Restormer is investigated. An experimental dataset consisting of 90 images captured at various distances from 40 to 120 cm was utilized for evaluation. Experimental results indicate that Restormer tends to achieve lower NIQE values while maintaining competitive Tenengrad values under the present evaluation setting. It is concluded that the proposed computational framework is useful for LC-lens-based text imaging under the present experimental conditions.
Takewaki et al. (Thu,) studied this question.