Hilbert Huang Transform And Its Application In Feature Extraction

The Hilbert Huang transform which has data driveability can decompose signals into multiple intrinsic mode functions(IMFs)components based on the local characteristics of the signal.It reflects the local characteristics of non-stationary signals better so that the characteristics of non-stationary signal are extracted more accurately.The identification of chromatogram of traditional Chinese medicine is an important basis for the quality control of traditional Chinese medicine,the identification of true and false Chinese medicine and the cultivation of Chinese medicinal materials.Due to different origins or different cultivation methods,the chromatogram of the same traditional Chinese medicine exhibits certain difference,but the difference is marginal.The high similarity has brought difficulties to the identification and become the key problem of the quality control of traditional Chinese medicine,the identification of true and false Chinese medicine and the cultivation of Chinese medicinal materials.The key to solve this problem is to solve the problem of feature extraction of chromatogram of traditional Chinese medicine.The chromatogram of traditional Chinese medicine is a series of one-dimensional signals composed of Gaussian function of different frequencies.Its feature extraction should be a kind of nonstationary signal feature extraction problem.Illumination problem is one of the key problems in face recognition.The research of feature extraction is still the focus of face recognition.The digital face images are one kind of two-dimensional signals.The different frequency components caused by illumination variation increases the difficulty of face identification.These two kinds of pattern recognition problems may have different backgrounds and different modes,but they share the characteristic of non-stationary signals.Therefore,it is necessary and feasible to apply the Hilbert transform method to the chromatogram of traditional Chinese medicine and the illumination of face recognition.The main contents and novelties of the dissertation are as follows:1)Feature extraction of the chromatogram of Glycyrrhizae in different cultivation conditions by using Hilbert Huang Transform(HHT)This study will identify the same Chinese medicinal herbs,which were cultivated under different cultivation conditions.In this dissertation,the glycyrrhiza fingerprint of medicinal herb is considered as a signal sequence,and a new approach which is based on empirical mode decomposition(EMD)of HHT and fractal dimension is applied to analyze and construct the feature vector of a glycyrrhiza fingerprint of medicinal herb.The EMD fractal features are extracted through computing the fractal dimensions of each IMF.The novel approach is applied to the recognition of the three types of glycyrrhiza fingerprints.Experiments show that EMD fractal features achieve higher recognition rate than that of fractal-wavelet features in the case of concentration-change,i.e.the number of peak and peak drift of sample which has slight changes.Based on this,this dissertation presents a new approach called the empirical mode decomposition - window fractal(EMDWF)algorithm in classification of fingerprint of medicinal herbs.In this way,a glycyrrhiza fingerprint of medicinal herb is considered as a signal sequence,and the EMD and Hiaguchis fractal dimension are applied to construct a feature vector.By using EMD,the glycyrrhiza fingerprint of medicinal herb can be decomposed into some IMFs.As windows fractal dimension(WFD)applied to each IMF and original signal,the features of the glycyrrhiza fingerprint of medicinal herb can be obtained.Thereafter,SVM is applied as a classier.The results of the experiments demonstrate clearly that the feature extracted by EMDWF is better than that of the existing methods including the pure EMD.With the increase of the number of training samples and the increase of the number of layers in EMD,the classification result achieves more stability.2)High frequency feature extraction of the illumination face using HHTIn order to solve the problem of face recognition under different illumination conditions,this dissertation proposes a feature extraction method based on EMD high frequency IMF and feature extraction method with face fusion based on high frequency IMF.The first IMF is used as the feature of illumination face recognition.The experimental results on Pose,Illumination,and Expression(PIE)face database of the Carnegie Mellon University show that the recognition performance is better than that of db4 wavelet transform.On this basis,this dissertation proposes a feature extraction method with face fusion based on high frequency IMF.After the illumination face is fused by high frequency IMF,the influence of different directional light sources on the face image is eliminated.The experimental results on the PIE face database show that the recognition rate is higher than that before fusion.3)The mathematical analysis method is used to improve the numerical method of Hilbert Huang TransformEMD of Hilbert Huang Transform is a method of numerical calculation.This paper proposed improving two-dimensional moving average filter instead of the mean of upper and lower envelope,which was called bidimensional EMD(BEMD).BEMD decomposes illumination face images into a complete series of bidimensional Intrinsic Mode Functions(BIMFs)which capture the intrinsic frequency components of original signals.For each BIMF by Riesz transform,the corresponding monogenic signals were constructed.Phase Congruence(PC)contained phase information of images is calculated and has been adopted as facial features to classify different faces under variant illumination conditions.The experimental results show that the proposed method is better than thetraditional method.The results of the research not only have important significance for the feature extraction of the chromatogram of Glycyrrhizae and illumination face images but also play a beneficial role in feature extraction of nonstationary signal and theory of Hilbert Huang transform.