Restauração de imagens utilizando aprendizado de máquina

Abstract

Image processing is an area that has received considerable attention as a result of the evo- lution of digital computing technology. One of the main techniques of image processing concerns its restoration, which consists in smoothing noise and detail enhancement, which are altered due to problems in the process of forming and transmitting the image. Based on the efficacy of sparse techniques and machine learning found in literature in the context of image restoration, we propose the union of these techniques as well as their evaluation in grayscale images. We also propose a study of energy-based networks such as Restricted Boltzmann Machines for noise suppression in binary images and the application of newer classifiers in this context, such as Optimum-Path Forest. Experiments using a public data- base corrupted by different degradations such as noise and/or blurring show the ineffective application of sparsity to different neural network architectures, the effectiveness of the Restricted Boltzmann Machines and the Optimum-Path Forest classifier.