Image super-resolution (SR) is the process of artificially producing a high-resolution (HR) image from one or several low-resolution (LR) images. Image super-resolution techniques are based on interpolation, reconstruction and learning.
In recent times, learning-based super-resolution algorithms are widely used. In learning-based super-resolution algorithms, HR image is obtained from a single LR image using training database. In these algorithms, the priori information is derived from the training database.
Ro Mi Ha, a lecturer at the Faculty of Information Science and Technology, has proposed a learning-based image super-resolution for a single LR image using discrete wavelet transform (DWT) and Gaussian mixture model (GMM).
In this method, if a low resolution (LR) input image and a database consisting of low and high resolution images are given, a high-resolution image for the input image is obtained by learning of the high-frequency details from the database. Then, high-frequency details of an HR image are described as wavelet coefficients at finer scale using DWT. The conversion function for obtaining the finer wavelet coefficients of an HR image from the coarse wavelet coefficients of an LR image is set as a weighted linear transformation using GMM.
She has demonstrated the effectiveness of the proposed method by conducting some experiments on gray images.
The proposed method can be used in applications such as remote surveillance where the memory, the transmission bandwidth and the camera performance are the main constraints.
© 2021 Kim Chaek University of Technology