Two-dimensional EMD Method And Its Application In Image Processing

Image texture analysis is of important significance in the field of digital imageprocessing，it is widely used in remote sensing image processing，target recognition based ontexture，texture based image retrieval and other practical applications．Doctor NordenE．Huang creatively put forward the empirical mode decomposition algorithm in1998，at firstthe algorithm was mainly used in the multi-resolution analysis of one dimensional timeseriessignal，such as seismic signal analysis and data analysis of ocean ripple．Unlike previousFFT and wavelet transform method，the EMD method decomposes the signal into a series ofintrinsic mode function and a residual．These intrinsic mode functions satisfy orthogonalityand completeness，and they together with the residual can realize perfect reconstruction of theoriginal signal．In2003，Nunes proposed the bidimensional empirical mode decompositionwith strict sense，and used this method to analyze image texture．The key content of this paper is as follows.Firstly I described the basic principle ofone-dimensional empirical mode decomposition,and then introduced the two-dimensionalempirical mode decomposition algorithm.Also I analysed several key issues in thetwo-dimensional EMD decomposition process,and proposed an improved two-dimensionalEMD method with good performance in the selection of the extreme points and the boundaryeffect treatment,on the basis of the traditional two-dimensional empirical modedecomposition.Then I gave the concrete steps to achieve the improved algorithm,andcompleted the program code in Matlab environment.The actual Lena image and theCameraman image were decomposed at different scales，and the decomposition results wereanalyzed.Secondly Combined with the improved bidimensional EMD method and minimumdistance classifier，I achieved the effective classification of texture images．The textureimages were decomposed into a number of intrinsic mode functions and a residual image，andangular second moment，contrast，correlation，homogeneity，and entropy were calculated fromthe intrinsic mode function images.The five characteristics formed a feature vector to classifythe texture image.And according to the outstanding contributions of the texture feature vectorof IMF1and IMF2in the texture image classification,a new texture image classificationmethod was proposed based on the improved two-dimensional EMD method,and the correctclassification rate of this method can reach85.83%.Thirdly, the two-dimensional empirical mode decomposition method was used in image edge detection and image enhancement.Theimages were decomposed by the two-dimensional empirical mode decomposition method,andthe IMF at the smallest scale with binarization and morphological refinement was extracted toextract the edge information of the images.In the image enhancement，Each scale componentafter a two-dimensional EMD decomposition was enhanced respectively and thenreconstructed，and the enhancement method of each scale component used gray-scale lineartransformation and histogram equalization enhancement method,and finally obtained theenhanced image with ideal visual effects.In this paper，the image decomposition with bidimensional EMD method，the featureextraction of each component with gray level co-occurrence matrix，and the texture imageclassification with minimum distance classifier experiments were finished in the Matlab7.1programming environment，and the more accurate texture image classification results alsoillustrated the effectiveness of the bidimensional EMD method and rationality of the extractedfeature．In addition,the two-dimensional EMD method also obtained satisfactory results inimage edge detection and image enhancement,and the experimental results highlighted thesuperiority of the two-dimensional EMD method in the field of image processing.