China
Africa
Explore chapters and articles related to this topic
Registration for Super-Resolution: Theory, Algorithms, and Applications in Image and Mobile Video Enhancement
Published in Peyman Milanfar, Super-Resolution Imaging, 2017
Patrick Vandewalle, Luciano Sbaiz, Martin Vetterli
Super-resolution algorithms typically combine multiple aliased images with small relative motion, and create a single high-resolution image. The input can be a set of pictures taken with a digital camera from approximately the same point of view. An application could be to use a low-resolution camera (with a good optical system), capture a set of images while holding the camera manually in approximately the same position, and use the small movements of the camera to reconstruct a high-resolution image. This would allow to take multiple images with a cheap camera, and combine them to a higher resolution image as if it had been taken with a more expensive camera. Other applications can be found for example in situations where a camera sensor cannot be easily replaced, such as in satellites. It is (almost) impossible to install a new camera sensor, while a modification of the software enables a series of images of approximately the same subject to be taken.
Deep Learning for Retinal Analysis
Published in Ervin Sejdić, Tiago H. Falk, Signal Processing and Machine Learning for Biomedical Big Data, 2018
Henry A. Leopold, John S. Zelek, Vasudevan Lakshminarayanan
Super-resolution techniques are methods for taking low-resolution or low-quality images and enhancing them to high quality through a number of methods including multiresolution techniques and pixel-energy-based quality normalization functions [61]. Applying these methods to retinal images during initial preprocessing or downstream feature extraction steps can greatly improve the the system’s ability to detect and properly classify morphology [62]. A review of these techniques in retinal fundus applications can be found in Ref. [63] and nonretinal applications in Ref. [64]. Specific traditional techniques for analyzing different retinal morphologies are covered in subsequent sections alongside their deep learning counterparts.
Chemometrics and Super-Resolution at the Service of Nanoscience
Published in Klaus D. Sattler, st Century Nanoscience – A Handbook, 2020
Marc Offroy, Ludovic Duponchel
The super-resolution is defined by the use of image processing methods from the signal processing research field to overcome limitations of optical systems. The main idea of super-resolution concept is the fusion of several low-resolution (LR) images of the same “object” from slightly shifted point of view to generate one higher-resolution image i.e. with higher spatial details (Farsiu et al. 2004a). The N LR images (i.e. low pixel density) are defined by M1 × M2 pixels and are denoted {Ck}k=1N. These LR images can be considered worse and different representations of a single image with much-higher resolution noted CSR defined by L1 × L2 pixels, where L1 > M1 and L2 > M2 for 1 ≤ k ≤ N. The parameters L1 and L2 are user-defined that always satisfy the following rule of thumb: L1L2≤NM1M2
Super-resolution GANs for upscaling unplanned urban settlements from remote sensing satellite imagery – the case of Chinese urban village detection
Published in International Journal of Digital Earth, 2023
Alessandro Crivellari, Hong Wei, Chunzhu Wei, Yuhui Shi
The scope of super-resolution models is directed to recover high-resolution information from corresponding low-resolution imagery. In recent years, deep learning has been intensively exploited for super-resolution solutions, achieving state-of-the-art performances in a variety of applications (Yang et al. 2019). Our case aims to attempt on leveraging super-resolution processes to transform an acquired remote sensing data source into a different and higher-quality one. Specifically, we adopt a GAN-based training process to increase the resolution of a low-resolution data source to ideally match the one of a destination high-resolution source. In practice, given a low-resolution input image, the model is intended to estimate a super-resolved image, artificially-generated version of an ideal original high-resolution image, differing from the input one by a scale factor .
A new approach for super-resolution and classification applications on neonatal thermal images
Published in Quantitative InfraRed Thermography Journal, 2023
Fatih Mehmet Senalp, Murat Ceylan
Super-resolution is a method of creating another high resolution image (SR) using the observed low resolution images by minimising distortions caused by the process of image obtaining and increasing the high-frequency components [3]. In other words, it is an approach that aims to provide better-detailed information by increasing the number of pixels per unit area in an image. Super-resolution has become an active field of study as it overcomes the resolution constraints of imaging systems and improves the performance of many image processing applications. The basic methods used in the early studies on super-resolution are the nearest neighbour, bilinear and bicubic interpolation techniques. These methods are based on the neighbourhood relationship of pixels [4]. Here, bicubic interpolation provides better quality images than the other two methods, but it is insufficient for applications where edge detail information is very important (medical etc.). Thanks to hardware advances, image quality improvement works using deep learning-based architectures have become popular in recent years. It has been observed that deep network architectures achieve successful results in super-resolution applications. For this reason, many projects have been performed on super-resolution applications used in various fields [5]. In general application areas: recognition and detection studies using face, object, iris, eye, fingerprint, sign, licence plate and astronomy images etc. Also, quality enhancement of compressed video and images, remote sensing and biomedical image processing etc. example can be given [6–14].
Texture-driven super-resolution of ultrasound images using optimized deep learning model
Published in The Imaging Science Journal, 2023
Additionally, the dynamic artifacts brought on by the movement of sound waves during the scanning procedure also affect US images. Applying the appropriate image-enhancing techniques and increasing the input image resolution is essential for a greater classification rate [7]. The most popular technique for obtaining higher resolution (HR) images from lower resolution (LR) images is super-resolution (SR). Super-resolution is mostly used to improve the quality of a low-resolution image by utilizing a high-resolution counterpart. In order to create an output image that can disclose finer features that were not visible in any of the original low-resolution photographs, the super-resolution research first combined the information content of several low-resolution images.