Demoiring Images using Efficient Attention Fusion module in Wavelet Domain

Abstract

When we use a capturing device to capture digital displays, we obtain a vivid rainbow pattern. It's known as "moiré" and it has an impact on image quality and subsequent processing. There are several ways to save money. Modern approaches for eliminating moiré patterns rely on down sampling pooling layers to retrieve multiscale information, which might result in data loss. To show this issue, this research provides a wavelet-based demoiréing approach. Both Inverse Discrete Wavelet Transform (IDWT) and Discrete Wavelet Transform (DWT) are used in this. This technology may efficiently enhance the network related field without any data loss in place of standard down sampling and up sampling. Furthermore, this approach employs an effective attention fusion module to reconstruct additional details of moiré patterns (EAFM). Using an aggregate of green channel attention (ECA), spatial attention (SA), and neighborhood residual getting to know, this module can self-adaptively study several weights of function statistics at distinct degrees and encourage the community to cognizance extra on required statistics as moiré statistics to beautify the getting to know and demoiréing paintings performed via way of means of enhancing photograph excellent and offering the specified Accurate output. Extensive trials the usage of publicly to be had datasets have established that this approach can perpetually cast off moiré styles even as additionally being quantitatively and qualitatively sound.