基于局部纹理统计模型的纹理图像检索

提出了局部纹理统计模型与改进的Log-likelihood统计相结合的纹理图像检索方法。首先设计了3种不同角度分辨率和空间分辨率的＂Uniform＂模式的局部二值模式,并且用其来提取图像特征,然后建立图像局部二值模式和局部方差的联合统计模型,最后用改进的Log-likelihood统计进行图像检索。提出的基于局部纹理统计模型的纹理图像检索算法具有旋转不变性,克服了LBP算法对旋转比较敏感的缺点。通过对Brodatz图像库中50类600幅纹理图像进行仿真试验,新算法能得到89.71%的平均查准率,与基于Gabor小波＋WMV组合算法相比,新提出算法的查准率提高了19.05%。

Texture Image Retrieval Based on Local Texture Statistical Model

A novel texture image retrieval method based on local texture statistical model combining with improved Log-likelihood statistic was presented.First,image feature was obtained by "Uniform" local binary patterns of three different spatial resolutions and three different angular resolutions,and "Uniform" patterns were recognized to be a fundamental property of textures as they provide a vast majority of local texture patterns in examined textures,corresponding to texture microstructures.Second,joint statistical model between local binary pattern and local variance was built.Last,the experiment of texture image retrieval was done using the improved Log-likelihood statistic.The novel approach is more insensitive to rotation than the approach of LBP.Extensive experiments from Brodatz texture images database of 600 texture images containing 50 textures clearly show that the retrieval accuracy of new method can achieve 89.71%,which obtained 19.05 percent higher than that of the method based on the combination of Gabor wavelet with WMV.