A Deep Convolutional Autoencoder Architecture for Automatic Image Colorization

Producción científica: Capítulo del libro/informe/acta de congresoContribución a la conferenciarevisión exhaustiva

Resumen

The inherent complexity of image colorization has motivated computer scientists towards the development of algorithms capable of simplifying the image colorization process. Despite the numerous advancements yielded by these efforts, there are still some limitations regarding the resulting image quality and its similarity to the ground truth counterpart. This paper proposes and implements a deep convolutional autoencoder architecture that maximizes the image colorization performance on two different datasets, the Fruit-360 and Flickr-Faces-HQ. To this end, a modification of the VGG16 model and a custom deep CNN model were assembled to predict and portray colors on grayscale images. We obtained mean absolute and square error results under the 0.01% on both datasets, demonstrating the substantial similarity between the output image and its ground truth counterpart. Additionally, the peak signal-to-noise ratio results of 27.72 (Fruits-360) and 26.86 (Flickr-Faces-HQ) indicate that the image colorization process introduces a relatively low drop in image quality.

Idioma originalInglés
Título de la publicación alojada2022 IEEE Colombian Conference on Applications of Computational Intelligence, ColCACI 2022 - Proceedings
EditoresAlvaro David Orjuela-Canon
EditorialInstitute of Electrical and Electronics Engineers Inc.
ISBN (versión digital)9781665474702
DOI
EstadoPublicada - 2022
Evento2022 IEEE Colombian Conference on Applications of Computational Intelligence, ColCACI 2022 - Cali, Colombia
Duración: 27 jul. 202229 jul. 2022

Serie de la publicación

Nombre2022 IEEE Colombian Conference on Applications of Computational Intelligence, ColCACI 2022 - Proceedings

Conferencia

Conferencia2022 IEEE Colombian Conference on Applications of Computational Intelligence, ColCACI 2022
País/TerritorioColombia
CiudadCali
Período27/07/2229/07/22

Huella

Profundice en los temas de investigación de 'A Deep Convolutional Autoencoder Architecture for Automatic Image Colorization'. En conjunto forman una huella única.

Citar esto