Multi-Level Adaptive Wavelet Convolutional Bidirectional Network-Based Demosaicking and Denoising for Digital Color Images

Modern digital cameras capture real images sequentially, carrying out various image processing stages. Denoising and demosaicking are often the first two processes, with the former aiming to decrease sensor noise and the latter converting a sequence of light-intensity data into color images. A multi-level adaptive wavelet convolutional bidirectional network (ML-AW-CBiNet) was proposed in this research. The hybrid model of convolutional neural network and bi-long short-term memory with a correlation of color module performs the initial pre-processing stage. The CNN-BiLSTM minimizes denoising and demosaicking. The undesired sounds are eliminated, and the colors of the images are restored in the second stage using the Extended Triple-dimensional Discrete Wavelet Transform (E-TriDWT). After that, the features from the proposed model are extracted using the Improved Discrete Cosine Transform (IDCT). The peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) determine the outcomes. The proposed model achieved 37.44 of PSNR and 0.9516 of SSIM, the best among existing models. The results show that the suggested model has a clear advantage over all existing models when the proposed model is compared to existing models. This study employed the Kaggle dataset for testing and training purposes.