| 摘要 |
A method and architecture for face detection under varying light conditions, tilt levels, different color characteristics of faces in a digital image are disclosed. In one embodiment, skin pixels in the digital image are selected. Any unselected pixels are then classified into bright pixels and dark pixels. Skin pixels are then selected from the bright pixels. Skin pixels are further selected from the dark pixels. All the skin pixels are then grouped into windows. Any window having skin pixels that is more than or equal to a predetermined number of skin pixels is then declared as a skin window. Candidate face regions are then formed by grouping the skin windows. Boundaries of candidate face regions are then determined. Eye regions are then determined in the boundaries of candidate face regions. Each candidate face region is then declared as a face, if the eye regions are determined in it. |
