Journal Article

Brain Programming for the Evolution of an Artificial Dorsal Stream

Dozal, León, Hernández, Daniel E., Olague, Gustavo, Clemente, Eddie

2014

Overview

This work describes the application of a new strategy called brain programming for automating the design of visual attention (VA) models. Nowadays, a term known as cognitive vision coined within the computer vision and cognitive research communities has been introduced to delimitate the kind of computer vision systems that are robust, resilient and adaptable to the task at hand through the incorporation of cognitive abilities. In particular, visual attention is considered as a critical factor whose main goal is to establish a relationship between the different properties or features of the scene with the aim of selecting the most suitable aspects for the task at hand. This paper follows a main trend in cognitive computation where the visual pathway is modeled through a succession of levels or layers. Here, the VA task is defined with the idea that several areas of the brain are in charge of its functionality in a hierarchical way. To achieve such functionality, we propose that an artificial process, mimicking the natural counterpart, would be charged of looking for a set of complex operations using an optimization/search process. The idea is to include such operations within a VA model that will be evolved according to a specific task. The aim of the whole process is to provide with the best solutions among the space of possible visual attention programs (VAPs) for a given problem. In this way, the article presents a methodology for automating the design of VAPs. Therefore, the final design can be seen as a cognitive vision system that is engaged in a purposive goal-directed behavior. The results obtained on a well-known testbed confirm that the proposal is able to automatically design VAPs that outperform previous man-made systems developed by VA experts, while providing readable results through a set of mathematical and computational structures.

• Publisher: Springer US,Springer Nature B.V
• Subject: Artificial Intelligence / Biomedical and Life Sciences / Biomedicine / Brain / Brain research / Computation by Abstract Devices / Computational Biology/Bioinformatics / Computer vision / Design / Evolution / Neurosciences / Optimization / Robustness (mathematics) / Search process / Vision systems