Performance Analysis of Discrete Wavelet Transform Based First-order Statistical Texture Features for Hardwood Species Classification

A simple and efficient discrete wavelet transform (DWT) based first-order statistical (FOS) texture descriptor is proposed in this paper to accurately classify the microscopic images of hardwood species. Primarily, DWT decomposes each image up to 8 levels using selected Daubechies (db1- db10) wavelet as a decomposition filter. Subsequently, four FOS features, namely, mean, standard deviation, skewness and kurtosis are employed to obtain substantial signatures of these images at different levels. The db3 based FOS texture features has achieved 96.80% classification accuracy compared to 93.20% classification accuracy obtained by local binary pattern features using linear support vector machine (SVM) classifier.