用于纹理分类的多元旋转不变自回归模型

A MULTIVARIATE ROTATION-INVARIANT SIMULTANEOUS AUTO-REGRESSIVE MODEL FOR TEXTURE CLASSIFICATION

本文在Kashyap提出的CAR模型的基础上,发展了一种多元旋转不变自回归纹理模型(RISAR)。利用RISAR模型的参数作为纹理图象的旋转不变特征,进行纹理分类,分类精度相对于CAR模型有较大的提高;同时识别速度比SAR模型大大提高。如果把RISAR模型用于多分辨率自回归纹理模型中,分类精度可提高到100％。

In this paper, a new model called the multivariate rotation-invariant simultaneous auto-regressive (RISAR) model is developed based on R. L. Kashyap's circular auto-regressive (CAR) model. The parameters of RISAR model are used as the rotation-invariant features of the image texture. Some classification experiments are performed, which prove that the classification power of the multivariate RISAR model is much stronger than that of CAR model. By using RISAR model in multi-resolution simultaneous auto-regressive (MRSAR) model, an 100 percent of classification accuracy rate is achieved.