自适应的Haar型LBP纹理特征提取算法研究

在提取纹理图像的Haar型LBP特征中,人为设定的判断阈值主观性强、局部性差,导致提取的纹理细节和边缘模糊、纹理图像的局部性易被忽略。为此,提出了一种自适应的Haar型LBP纹理特征提取算法。该算法在二值化Haar型特征时引入高斯加权矩阵,以此获得客观、符合纹理图像局部特征的自适应判断阈值和Haar型LBP特征。实验结果表明,该算法能够有效地避免人为设定阈值对纹理特征的影响,可以准确地描述图像的纹理特征,Brodatz标准纹理库分类的正确率也得到了进一步的提高。

An adaptive Haar LBP texture feature extraction algorithm

Due to strong subjectivity and poor locality of the artificial setting judgment threshold, in the process of extracting the Haar local binary texture (LBP), the extracted texture details and edges are not clear and the locality of texture image may be ignored. Therefore, we propose an adaptive Haar local binary pattern texture feature extraction algorithm, in which the Gaussian weighted matrix is introduced when the Haar characteristic is binarized. Subsequently the adaptive judgment threshold and the Haar local binary pattern which are objective and conform to the locality of texture image can be extracted. Experimental results show that the proposed algorithm can effectively avoid the influence of the artificial judgment threshold on texture feature and accurately describe the texture feature of images. Besides, the classification accuracy for Brodatz texture datasets can also be further improved.