Reviving Visuals: A Deep Learning Approach to Image Restoration and Enhancement
Volume 19, Issue 1, 2025
Download| Author(s): |
Syed Muhammad Daniyal* Iqra University, Karachi, Pakistan, syed.daniyal@iqra.edu.pk Umer Hussain Qidwai FAST, NUCES, Karachi, Pakistan, k.237842@nu.edu.pk Faiza Latif Abbasi Iqra University, Karachi, Pakistan, Faiza.latif@iqra.edu.pk Abdul Khaliq Institute of Business Management, Karachi, Pakistan, khaliq@iobm.edu.pk Noman Bin Zahid Iqra University, Karachi, Pakistan, noman.zahid@iqra.edu.pk Mohsin Mubeen Abbasi Iqra University, Karachi, Pakistan, mohsin.abbasi@iqra.edu.pk |
|---|---|
| Abstract | In modern developments in image synthesis, neural networks are used to decode the random generation of latent code into high-quality images. These still cannot offer any easy way to alter to real image alteration. A type called model inversion tries to recover a latent code, which, when encoded back into an image, is similar to a target image. The approaches currently do not consider the possibility of picture modification on a semantic level in search of an accurate pixel. The study empowers an in-domain constrained least squares inversion (CLSI) approach that mixes a domain-traversed encoder with a domain-regularized optimizer to circumvent this. The way enables the neural network to rely on its innate knowledge to perform both flexible editing and image reconstruction without the need for re-training by embedding the inverted code within the latent space within the network. In this research, the effect of different encoder structures, initial point of inversion, and parameter spaces on the quality versus semantic editability trade-off is explored. The paper reveals the details about how neural networks accumulate the semantic properties in latent spaces. Further, the research enhances the versatility of editing and enhancement of recent developments in image generation models. As far as we know, in-domain CLSI also has a significant potential to produce high-quality images and support semantic edits that will make sense (both of which will improve the real-world image edits). |
| Keywords | Constrained Least Squares Inversion (CLSI), Semantic Image Editing, Latent Space Inversion, Image Reconstruction |
| Year | 2025 |
| Volume | 19 |
| Issue | 1 |
| Type | Research paper, manuscript, article |
| Journal Name | Journal of Information & Communication Technology | Publisher Name | ILMA University | Jel Classification | - | DOI | - | ISSN no (E, Electronic) | 2075-7239 | ISSN no (P, Print) | 2415-0169 | Country | Pakistan | City | Karachi | Institution Type | University | Journal Type | Open Access | Manuscript Processing | Blind Peer Reviewed | Format | Paper Link | https://jict.ilmauniversity.edu.pk/journal/jict/19.1/3.pdf | Page | 14-20 |