A biological perspective of Viola-Jones face detection

Human face detection in digital images and videos is a mature technology, yet its operational performance is generally sub-optimal. Any improvement would be beneficial in many applications. Some computer vision approaches to object recognition, including face detection, have begun to achieve impressive levels of accuracy and robustness, yet lack a clear connection to known cortical constructs. This motivates investigations of biologically-inspired techniques. The mechanisms by which contour shapes, and in particular faces, are represented in cortex and the means that neural models and computer vision algorithms can more closely approximate these are examined. The OpenCV library implements a standard face detection solution, the Viola-Jones detector. A hybrid framework, with a truncated Viola-Jones cascade followed by neural population models, is considered in this current work.