Texture Feature Extraction Algorithm And Its Application In Object-oriented Classification Techniques Applied Research

Texture is one of the important distinctive feature that image shows in the local region. The effect of image classification relies heavily on the methods of extracting those textural features from images. Based on the traditional texture extracting methods, this paper proposed two more methods: the method based on the gray level-maximum variation co-occurrence matrix and the method based on the wavelet variation matrix. Both of the methods are applied in the object-oriented image classification techniques. The main research focus of this dissertation is given as follows:1. The texture extracting method that makes use of the gray level-maximum variation co-occurrence matrix is based on the statistical method in texture analysis. First we construct the maximum variation matrix that measures the intensities of the texture of the images, combine the gray level matrix and maximum variation matrix to the gray level -maximum variation co-occurrence matrix. Then we compute the statistics of the matrix as the features for the texture and organize the features into the vector. Make use of the texture image in the standard texture image database Brodatz, it has been proved that it is more effective and much simpler to use the method of the gray level-maximum variation co-occurrence matrix than the ordinary method of gray level-gradient co-occurrence matrix and gray level-Primitive Co-occurrence Matrix in the classification experiments.2. The method that is based on the wavelet variation matrix is the one that combines statistical method with the spectral method. First we segment the image in multiple scales using the multi-scale analysis of wavelet, then construct the texture variation matrix that reflects the directional properties and the intensities of the text image. Then we compute the statistics from the matrix as the image texture feature. rearrange the statistics into the feature vector. After applying the method of wavelet variation matrix, the non-sampling method of extracting the texture and the method of wavelet scale co-occurrence matrix in the experiments of classifying standard texture image of Brodatz, it reveals that the method based on the wavelet variation matrix is superior to the other two methods. 3. Very impressive results have been reached by conducting classification experiments using real aerial images and applying the Grey level-maximum variation co-occurrence matrix method and the method that is based on the wavelet variation matrix in the object-oriented classification techniques. In order to further improve the precision of the classification, we design the"voting rules"to determine the category.