Adaptive skin detection system on DTV

We propose a skin detection algorithmic system for a consequent video sequence which is fast, reliable and applicable. We adopted YCbCr color space which is competitive in extracting skin region from digital television (DTV) related images. Based on Gaussian color probability density and assumption that the color is homogeneous in a continuous scene of DTV signal, we adopted face detection part to pursue skin space locus from a scene to another scene so that adequate skin space is renewed by face detection. We target this algorithm to DTV applications. As speed of algorithm is crucial in real time system, we adopted motion estimation (ME) to find exact frames which need skin space to be updated due to scene change or skin color change. The core of this work is a practical adaptive skin space search algorithm meanwhile, reducing distortion from various image sources. Our adaptive skin locus update (ASLU) algorithm shows outstanding result comparing to static skin region detection algorithms'. Moreover, skin color is modeled on 2-dimensional CbCr space by imposing normal density modeling. Therefore, we determined the region of skin color existing in real, closest boundary to skin color and obscure region in the space. All of mentioned region will be processed by ASLU for homogeneous skin re-segmentation.