DnSAD: Defocus and Spatially Adaptive Deblurring for Mura Compensation in Display Panels

Abstract

Mura artifacts, characterized by spatially non-uniform brightness or color inconsistencies, are critical defects in high-resolution display panels. Although in-focus images are ideal for camera-based mura compensation, defocused images are commonly captured in practical inspection environments to suppress moiré artifacts. However, defocus blur combined with optical distortion introduces spatially varying blur, which complicates accurate compensation. In this paper, we propose a defocus and spatially adaptive deblurring (DnSAD) framework for mura compensation in display panels. DnSAD addresses this challenge by capturing an additional dot-pattern image that provides essential information about spatially varying point spread functions (PSFs). The proposed framework is built upon a deep Wiener deconvolution backbone, extended to support multiple PSFs. The spatial location of each PSF is encoded into the model via a spatial attention module, enabling localized restoration tailored to the spatial blur distribution. Furthermore, a multi-scale refinement strategy is employed to progressively reconstruct high-frequency details. Experimental results on both real and synthetic datasets demonstrate that DnSAD effectively restores fine-grained mura patterns while adapting to spatial blur variations, outperforming both conventional and recent learning-based approaches.