Big Skin Regions Detection for Adult Image Identification

摘要

In this work, we propose a hybrid big skin regions detection approach for adult image identification which is composed of color filtering, texture filtering and geometry filtering three parts. A subsection processing skin color model is employed to filtering the non-skin color pixels, and then a coarse degree based texture filtering is used to filter the pixels with rough texture and skin color tone, at last we filter the skin like region (such as deserts or beaches) by fractal dimension. The experimental results indicate this approach has high recall ratio and precision ratio.