Multi-scale Texture Analysis And Its Application On Remote Sensing Image Classification

A remote sensing image is a reflection of the scale dependent spatial structure and process. Remote sensing information at different spatial scales reflects the spatial differences between different structures at different scales. Image analysis at a single scale cannot reflect the heterogeneity and dynamics of the spatial structure and process.In this paper, two researches were carried out by applying multi-scale analysis method. One is high resolution remote sensing texture extraction method; the other is objected oriented classification method. Wavelet analysis was used in texture extraction algorithm. The multi-scale analysis in texture extraction algorithm refers to the scales of the wavelet decomposition. In the objected oriented classification, the concept of multi-scale does not refer to the resolution of the pixels but the image segmentation scales, which is the minimum threshold of the heterogeneity of the segmented image objects. Based on the concept of multi-scale, this paper focuses on the study of multi-scale texture analysis and object oriented classification theory and procedure. The main contents and conclusions are as below:1. A overview was given on the image texture analysis and classification methods related to multi-scale analysis. A texture extraction algorithm called Scale Co-occurrence Matrix Based on wavelet Transform was proposed and the extracted texture images were used for classification.2. Discussion was made on four different texture analysis methods, Statistical Method, Structural Method and Model Based Method. Comparison was made to conclude the advantage and disadvantage of these methods.3. Wavelet analysis can depict the multi-scale characteristics of the image and the grayscale co-occurrence matrix reflects the grayscale statistics of the image well. Hence, wavelet analysis and gray scale co-occurrence matrix were combined to extract texture information from high resolution remote sensing images. Scale Co-occurrence Matrix Based on Wavelet Transform was proposed in this paper. With the analysis of the textural features of different land-cover types, low frequencies and high frequencies images of the wavelet transformed results were used separately for texture extraction.4. Pixel based classification has the limitation of mixed pixels. Study was...