Determination Of Forest Texture Construction Factors Based On Glcm

The remote sensing images with very high spatial resolution have rich spatial information, which is more convenient to recognize the property features of the goal objects. It is a hot point to classify and recognize the objects using the texture information during the applying research of the RS images with the high resolution. Nowadays, there are four methods of the texture study basically: the structure method, the digital signals dissolving method, the model method and the statics method, and the last one is widely used. The Gray Level Co-occurrence of Matrix (GLCM) is the most popular way of the statics method, which relates to the construction factors such as the size of moving window ,the step, the gray level and the direction. The choice of the construction factors will affect the result of the method directly. However, within the researches now, it is only according to experiment or enumeration convincing the textural construction factors. Their weak points are: for one thing, it is difficult to achieve the best combination using the experiment; for anther thing, it will cost huge work of calculation to find the best construction factors.To solve the problem, we propose a good way to determinate the best or near the best textural combination quickly and exactly using the orthogonal experiment design technique. The orthogonal experiment design is a scientific and effective method to study and dissolve the muti-factor. It can obtain the reasonable and reliable result through the limited times of the experiment, which reduces a lot of calculating work, and improve the efficacy of the problem-untying. This article calculates texture images according to the 44 groups of the textural construction factors in the orthogonal table using 8 texture features, and constructs a separate index—J value, which can depict the different kinds of the forest textures according to the principle of the Separability Criterion based on the class to choose the best combination. With the result of the Multivariate Analysis of Variance, we can conclude that, in the WorldView RS image, the separability (J value) among the classes of the forest vegetation in the entropy texture image reaches top, and the best textural parameter combination is: the size of the window: 27×27, the step: 3, the gray level: 32, and the direction: 135°. During the UAV splice image, the separability (J value) in the correlation texture image is the largest, with the best textural parameter combination: the size of the window: 51×51, the step: 5, the gray level:256, and the direction: 0°.The result of this paper shows that: first of all, it is an effective way applying the orthogonal experiment design technique, which is put forward in this paper, to determinate the textural construction factors of GLCM and discriminate the forest texture; secondly, the Class Separability Criterion of the geometry distance can evaluate the function of the textur e images effectively.