The Research On Empirical Mode Decomposition And Its Application In Image Segmentation

As a special kind of image, texture image has the basic attribute of image as well as the typical characteristic of texture. Granularity, directionality and periodicity are features of texture. Texture segmentation is one of the important research subjects in image engineering. And texture image segmentation is also a difficult problem. Existing texture segmentation methods which base on wavelet, neural network, fractal theory, fuzzy theory and mathematics morphology have designed texture segmentation algorithms according the frequency characteristic and multi-scale characteristic. Texture segmentation models are formed according to priori knowledge and a certain unearthly idea ideology. Then, texture segmentation algorithm implements according the corresponding ideal model. However, priori model lack the adaptive characteristic. These algorithms hardly take into account intrinsic attribute and characteristic of problem.Empirical mode decomposition, which decomposes signal by analyzing signal's own characteristic, is proposed as a new signal analysis method in the middle of 1990s. Empirical mode decomposition is an adaptive multi-scale analysis method. It has adaptive characteristic. And it is regarded as the most radical breakthrough for the existing signal analysis methods. So the research on empirical mode decomposition is of great significance both in theory and in application.In this paper, one-dimensional empirical mode decomposition algorithm has been analyzed. This paper proposes a new border restraint algorithm and uses it in bidimensional empirical mode decomposition for two-dimensional image decomposition. A new fast bidimensional empirical mode decomposition algorithm is proposed.Then, this algorithm is improved. A new multi-scale, multi-directional bidimensional empirical mode decomposition algorithm is proposed. According to bidimensional empirical mode decomposition algorithm, a new texture segmentation algorithm is proposed. Experiment has shown that the border restraint algorithm properly restrains the border effect in the course of bidimensional empirical mode decomposition and decreases the calculation times. Bidimensional empirical mode decomposition which introduces new border restraint algorithm is not only faster but also more effective. Characteristic extraction which bases on bidimensional empirical mode decomposition is very accurate.Texture segmentation effect is very good.