Dynamic weight learning for RGB image demosaicking with a Bayer color filter array

Most snapshot color imaging devices adopt a single sensor with a Bayer color filter array (CFA), where only one RGB component is captured at each pixel, requiring demosaicking to recover full-color images. Existing methods range from fast interpolation to deep learning approaches with high reconstruction accuracy but often suffer from a trade-off between performance and computational complexity. To address this issue, we propose an efficient demosaicking method based on dynamic weight learning. The proposed network adaptively computes layer-wise feature weights without increasing model parameters and incorporates a mixed-attention block to jointly exploit global and local information, effectively reducing reconstruction artifacts. Extensive experiments demonstrate superior performance over that of existing methods.