A Non-Linear Convolution Network for Image Processing
Autor: | Stefano Marsi, Romina Soledad Molina, Giovanni Ramponi, Jhilik Bhattacharya |
---|---|
Přispěvatelé: | Marsi, Stefano, Bhattacharya, Jhilik, Molina, ROMINA SOLEDAD, Ramponi, Giovanni |
Jazyk: | angličtina |
Rok vydání: | 2021 |
Předmět: |
Compression artifact
Computer Networks and Communications Computer science neural network single-image super-resolution edge-preserving smoothing lcsh:TK7800-8360 Image processing 02 engineering and technology Edge-preserving smoothing neural networks non-linear convolution adaptive filters noise removal image deblocking JPEG artifacts removal Transfer function Convolution 0202 electrical engineering electronic engineering information engineering Electrical and Electronic Engineering Artificial neural network lcsh:Electronics 020202 computer hardware & architecture Adaptive filter Hardware and Architecture Control and Systems Engineering Signal Processing 020201 artificial intelligence & image processing Algorithm Smoothing adaptive filter |
Zdroj: | Electronics, Vol 10, Iss 201, p 201 (2021) Electronics Volume 10 Issue 2 |
Popis: | This paper proposes a new neural network structure for image processing whose convolutional layers, instead of using kernels with fixed coefficients, use space-variant coefficients. The adoption of this strategy allows the system to adapt its behavior according to the spatial characteristics of the input data. This type of layers performs, as we demonstrate, a non-linear transfer function. The features generated by these layers, compared to the ones generated by canonical CNN layers, are more complex and more suitable to fit to the local characteristics of the images. Networks composed by these non-linear layers offer performance comparable with or superior to the ones which use canonical Convolutional Networks, using fewer layers and a significantly lower number of features. Several applications of these newly conceived networks to classical image-processing problems are analyzed. In particular, we consider: Single-Image Super-Resolution (SISR), Edge-Preserving Smoothing (EPS), Noise Removal (NR), and JPEG artifacts removal (JAR). |
Databáze: | OpenAIRE |
Externí odkaz: |