Wood board image processing based on dual-tree complex wavelet feature selection and compressed sensing

Abstract

In order to identify wood surface types and carry out board grading, a collaborative classification method without image segmentation is proposed to sort four wood surface types: radial texture, tangential texture, live knot and dead knot. Firstly, three-level dual-tree complex wavelet decomposition is applied to the board image and 40-dimensional feature vector is obtained; secondly, particle swarm optimization (PSO) algorithm is used for feature selection and dimension reduction, and after optimization 11 key features are chosen; finally, the surface types are identified by compressed sensing classifier based on the features optimized by PSO. Four types of Xylosma samples are used for the experiment. The classification accuracy of the above four types are 100, 86.7, 96.7 and 86.7 %, respectively. Theoretical and experimental results show that the direction property of dual-tree complex wavelet can express the complex information of wood surface and the classification realized by compressed sensing is effective.