An approach of bag-of-words based on visual attention model for pornographic images recognition in compressed domain

摘要: Bag-of-words (BoW) model has been widely used in pornographic images recognition and filtering. Most of existing methods create BoW from with a scale-invariant feature transform (SIFT) descriptor the pixel domain. These require extra processing time to decompress compressed formats. In addition, SIFT only views local points centers some regions as BoW, which ignores major role image region human visual system. Different above this paper, approach based on attention is proposed recognize domain, includes following steps: (1) face detected remove or ID photo benign images; (2) built according characteristics image; (3) are by domain; (4) four features color, texture, intensity skin extracted regions; (5) created k-means cluster (6) will be represent images. Experimental results show that can more accurately less computational time.