Sequential Attention and Saccadic Grouping for Object Discrimination Using Reinforcement Learning

An important issue in sequential object recognition is to de ne a strategy for saccadic access of visual information, and the representation of the features under observation. The original contribution of this work is to embed the saccadic decision procedure in a cascaded recognition process where visual evidence is probed exclusively at salient image locations. In a rst processing stage, salient image locations are determined from the local entropy in object discrimination. Local information in terms of code book vector responses contribute to the recognition state in the MDP. A reinforcement learner performs then trial and error search on useful actions towards salient locations within a neighborhood, receiving reward from entropy decreases. The method is evaluated in experiments on object recognition using the COIL-20 database, proving the method being computationally feasible and providing high recognition rates.