<p>Currently, the implementation of convolutional neural network(CNN) in the domain of single image super resolution(SISR) is developing gradually. Despite numerous CNN—based methods have accomplished splendid conduct, it yet struggle with the feature extraction problem for proper image reconstruction. We present an optimized and deep cross dense skip connected network for SISR (DCDSCN) which target upon feature extraction of LR image for better reconstruction. The network consists of eight ResNet blocks which is combination of skip connection and convolution layers to preserve fine grained details, these blocks is named as cross—connected dense in dense convolution block(CDDCB). As the name suggests the CDDCB block employs cross—connected convolution layers using dense skip connection which alleviate the gradient vanishing problem. The <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="42979_2025_4006_Article_IEq1.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="39" /> </InlineMediaObject> <EquationSource Format="TEX">$1\times 1$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>1</mn> <mo>×</mo> <mn>1</mn> </mrow> </math></EquationSource> </InlineEquation> convolution layer refine the process of feature extraction while the <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="42979_2025_4006_Article_IEq2.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="39" /> </InlineMediaObject> <EquationSource Format="TEX">$3\times 3$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>3</mn> <mo>×</mo> <mn>3</mn> </mrow> </math></EquationSource> </InlineEquation> convolution layer in the lower branch allows the model to capture more spatial context thus this approach balances the overall model in terms of it’s performance and efficiency Moreover these CDDCB block incorporates dense skip connection within all blocks in order to back propagate gradient efficiently while making short paths between low resolution(LR) and high resolution(HR) image method hence titled as DCDSCN. The proposed architecture trades off well when compared with other existing SISR methods.</p>

Optimized and Deep Cross Dense Skip Connected Network for SISR (DCDSCN)

期刊论文
发表日期：

2025年5月26日

查看全文

Varsha Singh,
Naresh Vedhamuru,
R. Malmathanraj,
P. Palanisamy

摘要

Currently, the implementation of convolutional neural network(CNN) in the domain of single image super resolution(SISR) is developing gradually. Despite numerous CNN—based methods have accomplished splendid conduct, it yet struggle with the feature extraction problem for proper image reconstruction. We present an optimized and deep cross dense skip connected network for SISR (DCDSCN) which target upon feature extraction of LR image for better reconstruction. The network consists of eight ResNet blocks which is combination of skip connection and convolution layers to preserve fine grained details, these blocks is named as cross—connected dense in dense convolution block(CDDCB). As the name suggests the CDDCB block employs cross—connected convolution layers using dense skip connection which alleviate the gradient vanishing problem. The $1\times 1$ $1 \times 1$ convolution layer refine the process of feature extraction while the $3\times 3$ $3 \times 3$ convolution layer in the lower branch allows the model to capture more spatial context thus this approach balances the overall model in terms of it’s performance and efficiency Moreover these CDDCB block incorporates dense skip connection within all blocks in order to back propagate gradient efficiently while making short paths between low resolution(LR) and high resolution(HR) image method hence titled as DCDSCN. The proposed architecture trades off well when compared with other existing SISR methods.