Research For BEMD Method And Its Application In Image Processing

Recently, with the development of computer technology and signal processing technology, the image processing includes such branches as image segmentation, edge detection, texture analysis. Empirical Mode Decomposition (EMD) is part of Hilbert-Huang Transform, as a new signal processing method, it has obtained great progress in the non-linear, non-stationary signal processing and which shows a strong advantage and unique characteristic. It decomposes a complex non-stationary signal into several single component stationary signals in several different scales and a trend residual term, so this method has advantages of adaptability, stationarity and locality.In consideration of the successful application of EMD in various fields, it has been developed to the field of two-dimensional signal analysis and has obtained some certain development. But because of complexity of signal and limitations of some methods of bidimensional data, Bidimensional Empirical Mode Decomposition (BEMD) has some problem in signal analysis and processing accuracy, which is the focus of this thesis to study and improve.In this paper, bidimensional image is processed using MATLAB on the foundation of BEMD. After the deep study to the traditional decomposition method, it is found that there are some defects and places can be improved such as the boundary effect caused by surface envelope, the stop condition of component decomposition, termination condition of decomposition and feasible improvement Methods has been raised point at those problems. Mathematical statistics method has been used in this paper to observe the change regulation of the number and amplitude of extreme points, and the new stop condition is set, the improved BEMD is formed at last. The speed and precision of the traditional BEMD is improved, and the new BEMD has been applied and analyzed in the image processing. As a specific two-dimensional signal, the image has the information of non-stationary and multi-scale. A series of several image components which using the extreme point as scale and whose frequency is from small to large and residual component who represents the variation trend of image dissection can be got after component decomposition by BEMD. The image components provide the basis for Image denoising, image edge detection, image compression.