Recognition of facial sketch styles

Abstract The style recognition of facial sketches drawn by artists contributes great to the painting authentication, digital entertainment and law enforcement. This paper presents a framework to automatically classify different styles of facial sketches based on support vector machines (SVM) and selective ensemble (SE) strategy. Some handwritten features are deployed to feed into SVM classifiers. The framework proceeds as follows: firstly, each geometrically normalized image is divided into five important parts: the whole image, left eye, right eye, nose, and mouth; Secondly, based on the analysis to the technique of the facial sketch artists, gray histogram, gray moment, speeded up robust feature and multiscale local binary patterns are explored to extract handwritten features from each part; Thirdly, SVM is explored to learn the mapping relationship from handwritten features of each part to the style of the artist and thus we obtain multiple classification scores; Finally, we combine these complementary classification scores via SE scheme. Our model is able to achieve the recognition rates of 94% and 96% respectively on two groups of sketches drawn by five artists which would be available on the website.