A Novel Algorithm based on Contourlet Transform for Extracting Paint Features to Determine Drawing Style and Authorship

Abstract

Objective: To develop a hybrid authentication method, Painting Authentication using Contourelet Transform (PAUCT), combining a contourlet transform algorithm with HMT-Fisher distance information for the purpose of art authentication based on the analysis of the background of paintings. Methods/Statistical Analysis: Methodology includes feature extraction from samples, as well as modeling using Hidden Markov tree and Fisher distance information. This is followed by validation against the work of the original artist through feature testing, with final output measured and validated using a variety of statistical methods to determine accuracy. Findings: Application/Improvements: The proposed model improves accuracy in detecting fake art, to 85% from 80% in current works, due to its applicability to discrete data which allows brushstroke analysis at different resolutions.