中文摘要：

为解决LDP算子运算速度较慢、对于有些模式无法精确区分等问题，提出局部方向三值模式纹理描述子。采用自适应阈值，对像素8个方向边缘响应进行三值编码，提取并统计正边缘模式和负边缘模式，联合起来构成局部方向三值模式，作为最终特征来描述图像。在Brodatz数据集和CUReT数据集上将该模式与传统的LBP、LDP、FLDP、LTP算法进行对比实验，对比结果表明，与LDP算法相比，该模式在上述两个数据集上的分类准确率分别提高了5．42％和8．43％，运算速度提高了8倍以上；与LBP、FLDP、FTP相比，该模式也有明显优势。

英文摘要：

To solve some problems of LDP such as low running speed and failures on distinguishing some patterns,a descriptor called local directional ternary patterns （LDTP）was proposed.To extract LDTP feature,self-adaptive thresholds were adopted to code 8-directional edge responses into 3 patterns.The positive patterns and the negative patterns were extracted respectively and their statistical histograms were concatenated as the final feature of the original image.The final feature was called LDTP since the edge responses of different directions were coded into 3 patterns.The proposed descriptor was demonstrated in texture classification on Brodatz dataset and CUReT dataset compared with LBP,LDP,FLDP and LTP method.Experimental results show that the LDTP descriptor surpasses LDP in the classification accuracy on the mentioned datasets by 5. 42%and 8. 43%re-spectively,and the running speed of LDTP descriptor is 7 times higher than that of LDP.And it also has obvious improvements compared with LBP,FLDP and LTP.