A practical classifier for photographs and non-photographic images based on local visual features
Kei Terayama, Hirohisa Hioki · 2015
Classification of digital images into photographs and various kinds of non-photographic images has not been sufficiently studied but has many applications such as retrieval of real scene photographs from web sites and image databases. In this paper, we show that the combination of Bag of Visual Words of SURF features and histograms of LBPs for HSV and Luminance components (SURF+LBP(HSVL)) is simple, but works well as visual features for photographs and non-photographic image classification. We found that a classifier trained with SURF+LBP(HSVL) was the best among all the classifiers we tested using various visual features. Our classifier attained an accuracy of 96.8% for our image dataset and outperformed the other state-of-the-art classifiers.