Hierarchical Neural Learning for Object Recognition

We present a neural-based learning system for object recognition in still gray-scale images. The system comprises several hierarchical levels of increasing complexity modeling the feed-forward path of the ventral stream in the visual cortex. The system learns typical shape patterns of objects as these appear in images from experience alone without any prior labeling. Ascending in the hierarchy, spatial information about the exact origin of parts of the stimulus is systematically discarded while the shape-related object identity information is preserved, resulting in strong compression of the original image data. On the highest level of the hierarchy, the decision on the class of an object is taken by a linear classifier depending solely on the object´s shape. We train the system and the classifier on a publicly available natural image data set to test the learning capability and the influence of system parameters. The neural system performs respectably when recognizing objects in novel images.