Implementing Object-Based Attention in HMAX

📄 Project Abstract

In this report the biologically plausible model of object recognition in the primate visual cortex, HMAX, will be discussed and investigated with the intention of discovering whether it would be capable of supporting object-based attention. It is discovered that the feed-forward stream of HMAX produces feature vectors which contain more than enough information to classify oriented lines or simple shapes, but while the current feedback mechanism is found to produce convincing results for attending to oriented lines the same cannot be said of its performance on simple shapes. A heuristic was defined which for each object in a given scene would score the activation of the object relative to its size. The highest score was assumed to correspond to the choice the model had made and it was found to choose the correct oriented line for all cues, but for shapes would almost always choose a square. An attempt was made to improve the performance of the model by inspecting the component planes of...