Imagery Texture Analysis Based on Multi-feature Fractal Dimension

Texture is the important spatial structure information and primary feature of remote sensing imagery. It contains the surface information of imagery and the relationship with around environment. Firstly focusing on the basic principle of fractal, this paper adopts five different kinds of fractal algorithms to calculate the fractal dimensions of image texture towards the researching image, and gains five different fractal features. Secondly, according to the above obtained five fractal features, this paper makes the feature extraction under the principle of the transformation based on classibility criterion, and obtains two group new feature vectors. Finally, in the experiments, this paper carries on the image classification using the two new feature vectors, and gets the classification results. This paper adopts the fusion matrix to evaluate the precision of these classification results, and validates the feasibility, precision, and reliability of this method. 1. INTRODUCE Texture is the important spatial structure information and primary feature of remote sensing imagery. It contains the surface information of imagery and the relationship with around environment. As carrying on the classification toward RS imagery, texture feature is one kind of common classification standard and judgment principle. Among the various methods of describing texture feature, the fractal algorithm is better than other traditional algorithms in the aspects of anti-noise and attention the macrostructure and the microstructure of imagery. Fractal algorithm mainly expresses the extent of self-similarity and the feature of roughness. Simultaneously, this algorithm can also detect the abundance details of imagery, and then obtain the better classification effect. This paper focuses on the principles and techniques of image texture analysis, and validates the principles and methods with appropriate experiments. Firstly focusing to the basic principle of fractal, this paper adopts five different kinds of fractal algorithms to calculate the fractal dimensions of image texture towards the researching image, and gains five different images of fractal features. Secondly, according to the above obtained five fractal features, this paper makes the feature extraction under the principle of the transformation based on classibility criterion, and obtains new feature vectors in order to carry on the image classification. Finally, This paper adopts the fusion matrix to evaluate the precision of these classification results, and validates that the method proposed in this paper can obtain excellent classification results, and can guarantee the precision, the feasibility and the reliability of image classification. 2. FRACTAL FEATURE BASED ON THE TEXTURE OF IMAGE 2.1 Fractal Algorithm Based on The Discrete Fractal Brown Random Model The Fractal Brown Motion is one kind of typical mathematics model to describe the random fractal feature in nature. And its definition is (Dongsheng Wang, 1995): Set ( ) x Η Β is a random Gauss field, and 0<H<1. If the ( ) x Η Β satisfies this following formula: ( ) ( ) y x x x x y F < ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ Δ Β − Δ + Β Ρ = Η Η Η Η ) ( (1) Then we will call the ( ) x Η Β as the Fractal Brown Motion. If the x and x Δ in this formula are discrete values, then we can get the Discrete Fractal Brown Random field (abs. as DFBR field). Taking the gray values of image as the third dimension f(x, y) about the plane coordinates x and y, we can consider it as a discrete curving surface of gray, and this curving surface will satisfy the DFBR field partially. It can be expressed as: ( ) ( ) { } G y x y x f y x S ∈ = , | , , ,