Extracting subtle feature of target signal based on double tree complex wavelet transformation

Double tree after wavelet transform has good direction selectivity and invariant parallel movement. Based on analysing histogram of modulus sub-band corresponding to six high frequency sub-band after double tree wavelet decomposition, we put forward a new kind of subtle character, namely combination character of Lognormal distribution parameters and Gamma distribution parameter. Using the character to segment signal feature, and using edge smooth technology in the process of segmentation and using k-means clustering to realize unsupervised segmentation. Experiments show that the method of feature extraction is new, and edge accuracy of segmentation results and regional consistency has antinoise character, and it is a kind of effective subtle segmentation method.