A Local Modified U-net Architecture for Image Denoising
Latha H N and Rajiv R Sahay · International Journal for Modern Trends in Science and Technology · 2020
Image denoising is long well-studied issues in image processing, reconstruction and computer vision field, for a variety of image modelling problems.In this proposed work,an extension to traditional deep CNNs, Global skip connection, Normalization, Local pixels statistics, are included to obtain faster training and testing convergence of the model called Local Modified U-net (LM U-net).High frequency information of the pixels lost during decoding process is restored at the encoder end by processing local neighbor pixels.Results shown over60,000 training images, global skip connections providedgood improvements in feature learning for image restoration and denoising.The proposed LM U-net can also be effectively utilized for many other image restoration tasks similar to image super resolution, image deblurring KEYWORDS: U-net 1, Image Segmentation 2, DCNN 3, LM U-net 4, decoder and encoder 5.