Image super resolution based on residual dense CNN and guided filters
- PDF / 12,125,559 Bytes
- 19 Pages / 439.37 x 666.142 pts Page_size
- 54 Downloads / 180 Views
Image super resolution based on residual dense CNN and guided filters Mohammed Y. Abbass 1,2 & Ki-Chul Kwon 1 & Md. Shahinur Alam 1 & Yan-Ling Piao 1 & Kwon-Yeon Lee 3 & Nam Kim 1 Received: 3 May 2020 / Revised: 24 July 2020 / Accepted: 2 September 2020 # Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract
Convolutional neural networks (CNNs) have recently made impressive results for image super-resolution (SR). Our goal is to introduce a new image SR framework rely on a CNN. In this paper, the input image is decomposed into luminance channel and chromatic channels. A designed network based on a residual dense network is introduced to extract the hierarchical features from luminance part. The bicubic interpolation is simply used to upscale low resolution (LR) chromatic channels. However, this step degrades the chromatic channels. To tackle this issue, the SR reconstructed luminance channel is applied as the reference image in guided filters to promote the interpolated chromatic channels. Guided filters technique has ability to retain sharp edges and fine details from the reference image and carry them to the target images. Extensive experiments on several commonly used image SR testing datasets demonstrate that our framework has the ability to extract features and outperforms existing well-known techniques for image SR by LR image into the high resolution (HR) image efficiently. Keywords Image resolution . Bicubic interpolation . Guided filter . Sparse coding . Multi-scale deep super-resolution
1 Introduction Single-image super-resolution (SISR) is used to retrieve high-resolution (HR) images from input low-resolution (LR) images. This is demanded in digital image processing because LR images are obtained in a range of fields, including surveillance imaging, medical imaging, and satellite imaging [25, 26, 29]. SISR is an ill-defined problem because many pixels in HR
* Nam Kim [email protected] Extended author information available on the last page of the article
Multimedia Tools and Applications
images correspond to the same pixel in an LR image. Many SR methods have been developed based on key theories used to tackle the inverse problem, such as interpolation-, reconstruction, and learning-based methods. Interpolation-based approaches [27], including bicubic and bilinear approaches, depend on smoothness assumptions and are simple and fast. Reconstruction-based approaches [45] adopt LR images with regularizations to compute HR images. However, these approaches create visible blurring and undesirable aliasing along the salient edges, and they return awkward outputs on non-smooth patches because they contain poor high-frequency details. In recent years, many methods have been developed using learning-based SR approaches such as random forests [8] and sparse representation [46]. These approaches use datasets composed of LR and HR image patch pairs to extract mappings between LR and HR features. Tsai et al. first developed SISR in 1984 [37]. Since then, a range of SISR methods have be
Data Loading...
