A new generation of artificial intelligence contributing to improving the image quality
10.11591/ijaas.v15.i2.pp664-676
Salwa A. Alagha
,
Hadeel N. Abdullah
,
Suad Khairi Mohammed
High-resolution (HR) images provide inclusive and critical information, which is substantial for many implementations. Production operation for high-quality images from low-quality images can be costly and time consuming. The main advancement in this domain is produced by enhanced super-resolution generative adversarial network (ESRGAN); the ESRGAN and different deep learning (DL) models exhibit prominent advances in image super-quality. This research proposes introducing the discrete wavelet transform (DWT) as a multi-scale analysis stage that feeds into the network, whereby the frequencies are analyzed before being fed into the generative adversarial networks (GAN). The goal is to enhance the ability to recover edges and fine details, especially in low-resolution images. The performance of this proposed model is implemented, evaluated, and comparatively assessed. Key performance parameters, such as peak signal-to-noise ratio (PSNR) and structural similarity index metric (SSIM), are calculated, which compare the proposed model with other image-improving models (Bicubic, SRResNet, and ESRGAN). The experimental results indicate that the proposed method ESRGAN new yields a good result in image improvement, with a PSNR of (26.22, 26.00, 25.51, and 23.89) and an SSIM of (0.6638, 0.6255, 0.5882, and 0.6286) for four datasets, respectively.
