Features extraction based on wavelet packet transform for B-mode ultrasound liver images

This paper concentrates on the analysis of features extracted through the application of wavelet packet transform for B-mode ultrasound liver images. For years, the scientific community has been trying to figure out an effective method of feature extraction for revealing texture details and the succeeding classifications. Methods of feature extraction can be roughly classified into three categories: spatial-domain based, frequency-domain based and model based. Comparative study of typical methodologies within each category has proved that frequency-domain based algorithms, especially the wavelet transform are more effective in texture characterization and the succeeding classification. In our recent study, experiment result demonstrates wavelet packet transform is more advantageous than wavelet transform, intuitively because wavelet packet transform not only decompose the approximation of an image but also the details, thus can better describe texture surface. Single feature extracted will all go through feature selection procedure and finally the discriminative capacity of feature vector can be compared and verified by feature vector selection criteria and classification correct ratio of support vector machine.