An automated system for detecting human faces in generic scenes is reported in this dissertation. By inspiration from certain well established knowledge regarding the behaviors of some early stages in human visual pathways, the system makes its contributions to automated face detection in two aspects:; Functionally speaking, the system is able to overcome many of the difficulties imposed by this particular problem with far more ease and far less requirements on the input restrictions, as compared to most extant systems which employ computing techniques of all sorts but lack supports from the biological grounds. To be more specific, the proposed system is able to estimate coarsely the sizes of faces in the images first and then carry out the detection in a suitable scale space accordingly, disregarding the image qualities, variations in head numbers, poses and sizes, disturbances due to the existence of sun glasses, facial hairs and make-up, and even visual occlusion due to spatial arrangement in the scene. Mostly amazingly of all, the system uses no face models of any kinds and detects faces in a holistic style, i.e., faces in the scene simply get popped out from the background as a whole, with the contours naturally shaped out.; Technically speaking, the operations of the system essentially rely on the interactions among several types of visual information extracted by some functional modules emulating the behaviors of center-on and center-off ganglion cells and orientation selective cortex cells. The extraction is done by primitive operators working in either intensity or gradient space that are embedded in a process called bipolarized convolution, which is also responsible for the fusion, inhibition and excitation between visual information of different attributes. In short, a paradigm of multi-channeled visual information under a somewhat unified processing forms the backbone of the proposed system.; The effects achieved by the approaches undertaken in the proposed system, i.e., a paradigm of using massive primitive operations on appropriate kinds of information organized in proper forms in early stages of processing, may not necessarily be trivial, as suggested by the experimental outcomes, which are valid for face detection, and perhaps for other types of visual tasks that are still to be explored.
展开▼
