论文部分内容阅读
纹理是物质在外观上所呈现出的某种有规律的图案。反映在图像上,就是灰度的规律性分布。人们根据纹理特征可以区分大量的物质。利用计算机自动识别纹理,需要为纹理建立合适的数学模型。本文尝试用假设检验这个有力的数学工具对自回归纹理进行符合程度的检验。并运用这个工具来求得一个较合适的自回归模型。 本文对五种纹理的150个样本进行了检验,在0.01显著性水平下全部否定假设。因此可以得出结论:用自回归模型模拟纹理是比较恰当的。本文还提出了用假设检验手段来确定回归系数个数的方法,因而提高了识别精度,减少了运算量。本文提出的自相关系数代替自回归系数的方法进一步减少了运算量。用Fisher准则设计的分类器具有速度快、精度高、占用内存少等特点,对五类纹理抽取的自回归系数和自相关系数用Fisher分类器进行分类,分别取得了98.6％和97.3％的分类精度。 实践证明,纹理方法应用于图像识别是非常有效的,具有广泛的应用前景。
Texture is the appearance of the material in some kind of regular pattern. Reflected in the image, is the regular distribution of gray. People can distinguish between a large number of substances based on the texture features. Using a computer to automatically recognize textures requires creating a mathematical model of the textures. This paper attempts to test the conformity of autoregressive textures with this powerful mathematical tool. And use this tool to get a more appropriate autoregressive model. This paper examines 150 samples of five textures, all negatively assuming the 0.01 significance level. Therefore, it can be concluded that it is more appropriate to model the texture with an autoregressive model. In this paper, we also propose a method of using hypothesis testing to determine the number of regression coefficients, thus improving the recognition accuracy and reducing the computational complexity. The proposed autocorrelation coefficient instead of the autoregressive coefficient method further reduces the computational complexity. The classifier designed by Fisher criterion has the characteristics of high speed, high precision and less occupied memory. The autocorrelation coefficients and autocorrelation coefficients extracted from five types of textures are classified by Fisher classifier, and the classification is 98.6% and 97.3% respectively Accuracy. Practice has proved that the texture method applied to image recognition is very effective, with a wide range of applications.