Wood Materials Defects Detection Using Image Block Percentile Color Histogram and Eigenvector Texture Feature

Abstract

To automatic detect wood surface defects, a method based on image block percentile color histogram and eigenvector texture feature classification is proposed. Firstly, a wood surface image is divided into several same size image blocks. Secondly, for each image block, a percentile color histogram is calculated as image block color feature. Meanwhile, singular value decomposition (SVD) is adopted to extract k-max eigenvectors as image block texture feature. Then the percentile color histogram and eigenvector texture feature is combined to a feature vector for image block representation. Finally, a support vector machine (SVM) classifier is trained and used to determine which image block is sound or defect wood. The experimental results show that the proposed method can effectively detect wood surface defects, especially the knot type defects.