Efficient image filters that classify web-images as photograph of a real-scenes or as paintings are very desired in a content-based image retrieval system. The main contribution of this paper is the proposition of two feature vectors, the Receptive Field Profiles (RFPs) and the Composite Visual Feature (CVF), that effectively discriminate art-paintings from scene-photographs. The formulation of these signatures were inspired by the model and analysis of Human Visual System (HVS); The RFPs are approximated by a multi-channel color Gabor filters which capture color texture properties; The CVF measures color uniqueness, saturation and smoothness and edge discrepancy. Experimentations on a database of 20,000 images, collected from the web, with extremely variable visual contents, shown a very promising classification results (93%). We found that RFPs features are larger for photographs than for painting. The boundary separation between classes in feature spaces were modeled using Gaussian Mixture Models (GMM) and Support Vector Machines (SVM). A comparative analysis is conducted and GMM shown higher performance.