Advancing Image Deblurring Performance with Combined Autoencoder and Customized Hidden Layers

Abstract

This article introduces a novel approach to image deblurring by combining a Fourier autoencoder model. The proposed model effectively removes blur artifacts and restores image details by capturing frequency information using the Fourier Transform. In addition, the article presents a method to enhance deblurring by identifying optimal directions using an autoencoder model, trained on a dataset of blurry and sharp images to learn latent features for removing blur and restoring clarity. The encoded representations are used by the decoder to reconstruct a sharper version of the input image. A combination of two autoencoder models is employed, with a Convolutional Neural Network (CNN) handling the initial deblurring process and a fully connected model optimizing the deblurring parameters. This integrated approach leverages the strengths of CNNs in feature extraction and the flexibility of fully connected networks to produce higher quality, clearer images.