Extending IPOL to New Data Types and Machine-Learning Applications

Abstract

Image Processing On Line (IPOL) is a journal focused on mathematical descriptions of image processing (IP)/computer vision (CV) algorithms. Since the first article was published in 2010, it has started to become clear that the IP/CV discipline is mainly multi-disciplinary. For example, nowadays images are de-noised using convolutional neural networks (CNN), and fields such as neurophysiology need of the rudiments and techniques of IP/CV, general signal processing and artificial intelligence. IPOL wants to extend the capabilities of its demo system to cope with these needs. Specifically, in this article we review the state of the current demo system and its limitations. It enunciates a detailed project on how to build a more adapted system, and its minimal requirements: New data types, problematic heterogeneous data, the pre-processing and standardization, the possibility to chain different algorithms in a complex chain, and how to compare them.