Pornographic images detection using High-Level Semantic features

Abstract

The pornographic images recognition can be seen as a special kind of object recognition task,but current pornographic images filtering algorithms using BoVF approaches have some problems,such as the high false positive rate toward the bikinis images and insufficiency of filtering pornographic images with pornographic actions. The paper proposes a novel pornographic image filtering model using High-level Semantic features. Firstly, we optimize BoVW model to minimize semantic gap between low-level features and high-level semantic features and then high-level semantic dictionary is constructed by fusing the context of the visual vocabularies and spatial-related high-level semantic features of pornographic images. Experimental results show that the model has an accuracy up to 87.6% when testing the multi-person pornographic images, which is significantly higher than the existing pornographic images filtering algorithm based on Bag-Of-Visual-Words.