Image Texture Recognition By Wavelets

Along with the enhancement about the capability of computer and the demand in practice, the desire of image analysis is more and more strictly. Image texture recognition is the concernful content of the image analysis. Enhancing the accurate recognize rate is signality in practice. Wavelet analysis is a new subject in applied mathematics. We study texture recognition by wavelets in this article. Several results are obtained from the research of the image texture recognition. This study consists of four chapters, and the main contents of each chapter are as follows:In Chapter 1, we outline the basic knowledge of wavelet analysis that use in this study and research the development and current research situations of image texture recognition. The wavelet methods in image texture recognition are discussed, and put forward the necessity of the multiwavelet in image texture recognition.In Chapter 2, we investigate several methods in texture recognize. From the research of these methods, which indicate that choose the appropriate wavelet basis is the key of wavelet application.In Chapter 3, we investigate the choose rule of best wavelet basis in image texture recognition. The choice of best wavelet basis is an important problem of wavelet theory and it application. The investigation shows that the orthogonality, symmetrically and compact support of wavelet is relation to the analysis capable of the wavelet. Using the wavelet which has orthogonality, symmetry and compact support simultaneously can gain higher accurate recognition rate.In Chapter 4, we introduce the properties of multiwavelet from the orthogonality and approximation order. Because of the aliasing effect in practise, we introduce balance multiwavelet and multiwavelet pretreatment. Because the difference of the equipment and the condition of the collection, the same image may have different rotate and scale change. We construct wavelet arithmetic on the rotate and scale change texture recognition. We use the log-polar transform to eliminate the rotate and scale change, and gain row-shift log-polar image, and then give row-shift multiwavelet wavelet package arithmetic to eliminate the influence of row-shift and pick-up the energy signatures of the image efficiently. Experiments show that because the multiwavelet has orthogonality, symmetry, compact support and high approximation order simultaneously, the capacity of accurate recognition is better than others.