Texture image classification using improved image enhancement and adaptive SVM

Abstract

The development of a robust and accurate wood species recognition system based on wood texture images is important to guarantee the quality of the wood merchandise. Wood species can be classified according to distinctive texture features such as the positioning of pores or vessel, fibres, rays parenchyma, phloem, soft tissue, intercellular canals and latex traces. Since the quality of wood texture images obtained at the inspection site might not present at its optimum quality, blurry texture images captured during image acquisition process has been a challenging issue in designing accurate wood species recognition system. Therefore, a modified image enhancement method and an adaptive classifier are proposed in this study to overcome the above-mentioned problem. Firstly, an improved image enhancement method is proposed by fusing an unsharp masking with the conventional constrained least squares filter (CLSF) to enhance the blurry texture images. Secondly, an adaptive support vector machine is proposed for final classification. The wood texture images are classified based on the deep features extracted using a convolutional neural network model. The proposed system is also benchmarked with several image enhancement methods for comparison purposes. Investigation results proved that the proposed wood species recognition system is feasible in classifying blurry wood texture images.