A new era in computation
CVPRW '04 Proceedings of the 2004 Conference on Computer Vision and Pattern Recognition Workshop (CVPRW'04) Volume 12 - Volume 12
Robust Object Recognition with Cortex-Like Mechanisms
IEEE Transactions on Pattern Analysis and Machine Intelligence
Object Class Recognition and Localization Using Sparse Features with Limited Receptive Fields
International Journal of Computer Vision
LIBSVM: A library for support vector machines
ACM Transactions on Intelligent Systems and Technology (TIST)
Evolving a conspicuous point detector based on an artificial dorsal stream: SLAM system
Proceedings of the 14th annual conference on Genetic and evolutionary computation
Self-adjusting focus of attention by means of GP for improving a laser point detection system
Proceedings of the 15th annual conference on Genetic and evolutionary computation
Hi-index | 0.00 |
Computational neuroscience is a discipline devoted to the study of brain function from an information processing standpoint. The ventral stream, also known as the "what" pathway, is widely accepted as the model for processing the visual information related to object identification. This paper proposes to evolve a mathematical description of the ventral stream where key features are identified in order to simplify the whole information processing. The idea is to create an artificial ventral stream by evolving the structure through an evolutionary computing approach. In previous research, the "what" pathway is described as being composed of two main stages: the interest region detection and feature description. For both stages a set of operations were identified with the aim of simplifying the total computational cost by avoiding a number of costly operations that are normally executed in the template matching and bag of feature approaches. Therefore, instead of applying a set of previously learned patches, product of an off-line training process, the idea is to enforce a functional approach. Experiments were carried out with a standard database and the results show that instead of 1200 operations, the new model needs about 200 operations.