High-performance vision system exploiting key features of visual cortex

A computer method for recognizing objects in a subject image, comprising the steps of:

• providing a model that effectively replicates a visual cortex, the model executable by a digital processor and having a loose hierarchy of layers, each layer, from a lowest hierarchy level to a top level, providing increasing selectivity and invariance of an input image such that model output produces feature recognition and classification of an object in the input image;
• providing one or more bypass routes between layers, at least one bypass route from a layer at one level to a higher layer enabling increased selectivity and decreased invariance to the input image relative to a layer at a hierarchy level succeeding the one level, the at least one bypass route corresponding to a direct projection in the visual cortex;
• and
• rendering the model output, including displaying a recognized feature of the object and displaying an indication of a class of the object in the input image.

Computer apparatus for object detection and texture recognition, comprising:

• hierarchy means for providing a hierarchy of layers, corresponding to a visual cortex, each layer, from a lowest hierarchy level to a top hierarchy level, providing increasing selectivity and invariance of an input image such that a layer at the top hierarchy level produces detection of an object in the input image;
• bypass means for providing at least one bypass route between one layer and another layer higher than the one layer, such that object detection is generated from processing by the hierarchy layers from the lowest hierarchy level through the one layer to the other layer via the bypass route, and such that selectivity is increased and invariance is decreased at the other layer relative to a layer at a hierarchy level succeeding the one layer;
• windowing means for providing windows of input image data to the hierarchy means for determining shape-based objects in the input image;
• segmenting means for providing a segmented image of the input image to the hierarchy means for determining texture-based objects in the input image;
• and
• output means for displaying a recognized feature of the object and displaying an indication of a class of the object in the input image.