engleski

Automated Painter Recognition Based on Image Feature Extraction

This paper describes an approach to automated classification of paintings by artist. The individual style of an artist is recognized through specific elements of a painting which distinguishes the work of an individual from the works of others. The proposed method for automating the task of artist recognition focuses on the measurable elements in a painting which are represented with a set of global image features. The set of computed image descriptors includes statistical features that describe the intensity of a grayscale image, features based on color and textural features obtained using different techniques. Several classifiers were tested and their performance was evaluated on a collection of 500 digital captures of artistic paintings from 20 different artists, obtained from various Internet sources. Experimental results show overall classification accuracy of 75%. Comparison to the results of other painting classification approaches is difficult because this paper introduces several artists that have not been previously tested for automated painting classification, and due to the fact that image databases of existing classification methods vary highly in size and content.

image feature extraction; painter recognition; painting classification; visual art

nije evidentirano