A neural model combining attentional orienting to object recognition: preliminary explorations on the interplay between where and what

We propose a model of primate vision that integrates both an attentional orienting ("where") pathway and an object recognition ("what") pathway. The fast visual attention front-end rapidly selects the few most conspicuous image locations, and the slower object recognition back-end identifies objects at the selected locations. The model is applied to classical visual search tasks, consisting of finding a specific target among an array of distracting visual patterns (e.g., a circle among many squares). The encouraging results obtained, in which substantial speedup is achieved by the combined attention recognition model while maintaining good recognition performance compared to an exhaustive search, suggest that the biologically-inspired architecture proposed represents an efficient solution to the difficult problem of rapid scene analysis.