Adult image detection with c lose-up face classification

The SVM (support vector machine) and the SCM (skin color model) are used for detection of adult contents. The SVM consists of multi-class learning model and is very effective method for face detection, but complex. On the contrary, the SCM is very simple for detecting the adult images using skin ratio derived from statistical characteristics of RGB color information, but less effective in close-up facial images. So, we propose a hybrid scheme that combines the SVM for the 1^st filtering scheme using 3-class learning model (with classes of objectionable, non-objectionable and close-up facial image) with the SCM for the 2^nd filtering scheme using skin ratio. The performance of proposed scheme improves about 2.5% ~ 4.6% in the true positive rate and about 4.6% in the false positive rate.