A Novel Nonlinear Dimensionality Reduction Method for Robust Wood Image Recognition

The grading of woods is mainly determined by the defects on wood surfaces and determines the potential uses and values for the sawmills. However, the dimensions of wood images are high, which is difficult to deal with. Dimensionality reduction is one of key interests in processing the higher-dimensional image data without losing intrinsic information. The problem of sub-pattern based discriminative non-linear dimensionality reduction called Sp-DNDR is considered for wood image recognition. This setting uses the sub-pattern of the original samples data and within-class and between-class scatters are used to specify whether pairs of instances belong to the same class or not. Sp-DNDR can project the data onto a set of dasiausefulpsila features and preserve the structure of the data as well as the scatters defined in the feature spaces. We demonstrate the practical usefulness and high scalability of the Sp-DNDR for wood knot defects recognition tasks by extensive simulation experiments. Experimental results show Sp-DNDR based recognition method can achieve a higher accuracy. For dimensionality reduction, Sp-DNDR method outperforms some established typical dimensionality reduction methods. Besides, the proposed method has better robust to the interferences on wood surfaces.