Improving Image Filters with Cartesian Genetic Programming

Julien Biau, Dennis Wilson, Sylvain Cussat-Blanc, Hervé Luga

2021

Abstract

The automatic construction of an image filter is a difficult task for which many recent machine learning methods have been proposed. However, these approaches, such as deep learning, do not allow for the filter to be understood, and they often replace existing filters designed by human engineers without building on this expertise. Genetic improvement offers an alternative approach to construct understandable image filter programs and to build them by improving existing systems. In this paper, we propose a method for genetic improvement of image filters using Cartesian Genetic Programming. We introduce two operators for genetic improvement which allow insertion and deletion of a node in the graph in order to quickly improve a given filter. These new operators are tested in three different datasets starting from published or engineered filters. We show that insertion and deletion operators improve the performance of CGP to produce newly adapted filters.

Paper Citation

in Harvard Style

Biau J., Wilson D., Cussat-Blanc S. and Luga H. (2021). Improving Image Filters with Cartesian Genetic Programming. In Proceedings of the 13th International Joint Conference on Computational Intelligence (IJCCI 2021) - Volume 1: ECTA; ISBN 978-989-758-534-0, SciTePress, pages 17-27. DOI: 10.5220/0010640000003063

in Bibtex Style

@conference{ijcci21,
author={Julien Biau and Dennis Wilson and Sylvain Cussat-Blanc and Hervé Luga},
title={Improving Image Filters with Cartesian Genetic Programming},
booktitle={Proceedings of the 13th International Joint Conference on Computational Intelligence (IJCCI 2021) - Volume 1: ECTA},
year={2021},
pages={17-27},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010640000003063},
isbn={978-989-758-534-0},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 13th International Joint Conference on Computational Intelligence (IJCCI 2021) - Volume 1: ECTA
TI - Improving Image Filters with Cartesian Genetic Programming
SN - 978-989-758-534-0
AU - Biau J.
AU - Wilson D.
AU - Cussat-Blanc S.
AU - Luga H.
PY - 2021
SP - 17
EP - 27
DO - 10.5220/0010640000003063
PB - SciTePress