Multi-classifier Scheme with Low-Level Visual Feature for Adult Image Classification

As the usage and accessing of children to the web resources with porn images contain is growing, requirement of methods with high accuracy to detect and block adult images is a necessity. In this paper, a novel multi-classifier scheme is proposed based on low-level feature to exploit of advantages in classifier ensemble for achieving better accuracy compared to single classifier that applied to adult images detection. Low-level features are three different MPEG-7 descriptors include Color Layout Descriptor (CLD), Scalable Color Descriptor (SCD) and Edge Histogram Descriptor (EHD). In the classification part Support Vector Machine (SVM) and AdaBoost are applied and combined. Experimental results indicate that proposed scheme works better than each single classifier that used in the experiments.