Locating theWood Defects with Typical Features and SVM

Abstract

Local Binary Pattern (LBP) can provide us with the spatial structure of images and describe the original features better, such as curly edges, etc. And it has better luminance adaptability. Dual-Tree Complex Wavelet Transform (DTCWT) can maintain better time-frequency localized characteristics and extract the energy based statistical features, maintaining the limited data redundancy and effectual computation efficiency. Furthermore, the values of a board have a direct relationship with the grading determined by the defects on wood surfaces and determine the potential uses and the values for the Sawmills. In this paper, we effectively integrated the features by LBP and DTCWT to get the typical features for recognition. We proposed a wood defects recognition algorithm, which can effectively decrease the experimental errors and have better robust to the interferences. The method has been tested with color wood images. Based on visual valuation, the errors are relatively lower. After many comparative experiments, the results show the system is effectual and practical with better research values and potential applications.