Abstract: A new recognition method using two characteristics for image texture is proposed.In this approach，image texture is represented as a sequence of codes based on encoded contour lines for its texture patterns.By using suffix array，the characters of the texture are extracted from the chain strings based on MDL criterion，and its repeated frequency is calculated by flexible pattern matching algorithms in strings.Finally，different textures are recognized by the character strings and its frequency based on Bayesian clustering.Experimental results by using ore material photos demonstrate the effectiveness of the approach.

Key words: texture patterns, included angle chain, pattern recognition

摘要： 提出了一种能够结合纹理基元形状和大小两种特征的纹理识别方法。该方法首先定义一种编码规则，应用这种规则对图像中纹理基元的轮廓线进行编码，把纹理的表述转化为夹角链码形式。使用后缀数组提取链码中的重复字符串作为特征编码，应用动态规划算法在近似匹配条件下计算特征编码在链码中的重复频率。最后，以特征编码及其重复频率作为目标的形状特征和尺度特征，使用贝叶斯分类方法对纹理进行识别。通过识别实际矿石图片验证了这种方法的有效性。

关键词: 纹理基元, 夹角链码, 模式识别