Comparative Performance of Different Chrominance Spaces for Color Segmentation and Detection of Human Faces in Complex Scene Images

摘要: Color is a powerful fundamental cue that can be used at an early stage to detect objects in complex scene images. This paper presents analysis of the performance nine different chrominance spaces specific problem automatically detecting and locating human faces twodimensional still For each space, we use skin color model based on Mahalanobis metric segment from background by thresholding. We perform feature extraction segmented images fully translation-, scaleand in-plane rotation-invariant moments are derived Fourier-Mellin transform, apply multilayer perceptron neural network with invariant as input vector distinguish distractors. show for detection efficiency critically dependent goodness fit distribution proposed model, lesser extent discriminability between "non-skin" distributions. Also, normalized shown yield best segmentation results, subsequently highest rate large variety poses against relatively