Feature Extraction of Mold Defects on Fine Arts Painting using Derivative Oriented Thresholding

Abstract

Paintings can be damaged by natural causes or accidents. One of the crucial natural damages was frequently caused by mold defects. The mold discovery is an important step in the restoration of damaged paintings. The procedure is usually tedious and depends heavily on the qualitative visual judgement of an expert restorer. The aim of this work is to assist the restoration process via an automatic mold defect detection technique based on derivative and image analysis. This new method, designated as Derivative Level Thresholding (DLT), combines binarization and detection algorithms to detect mold rapidly and accurately from scanned high-resolution images of a painting. This work also benchmarks the performance of the proposed method to existing binarization techniques of Otsu’s Thresholding Method, Minimum Error Thresholding (MET) and Contrast Adjusted Thresholding Method. Experimental results from the analysis of 20 samples from high-resolution scans of 2 mold-stained painting have shown that the DLT method is the most robust with the highest sensitivity rate of 84.73% and 68.40% accuracy.