Figueroa Rosado, Juan

Publication Search Results

Now showing 1 - 1 of 1

Deep merge-and-run convolutional neural network for image denoising and super-resolution
(2024-05-10) Figueroa Rosado, Juan; Arzuaga, Emmanuel; College of Engineering; Sierra, Heidy; Rodríguez-Solís, Rafael A.; Department of Electrical and Computer Engineering; Florez Gomez, Edwin
Imaging systems have become part of ordinary life due to their integration into smart devices such as phones, tablets, and computers. In the field of image processing, we face a common challenge: noisy and low-resolution images. There are limitations and inherent problems to optical and imaging systems. Some of these are caused by noise, lighting, or vibrations. Recent years have seen significant advances in deep learning and machine learning techniques have been aimed at addressing these complex problems. This thesis introduces an innovative deep learning architecture composed of convolutional blocks in a merge-and-run configuration. The model tackles two prevalent problems in image processing: image denoising and single-image super-resolution. Our focus includes the design, implementation, and evaluation of this architecture, specifically targeting the denoising of confocal microscopy images and super-resolution of synthetic aperture radar (SAR) images. The model achieved highly competitive results in both use cases, enhancing image clarity and resolution comparably to existing methods, but with a 45% reduction in architecture size.