融合颜色-纹理模型的均值漂移分割算法

Improved Mean-Shift segmentation algorithm combining with color-texture pattern

针对常规的均值漂移算法在特征空间聚类时未考虑图像的纹理信息从而导致分割精度不高的问题,该文提出了一种融合颜色-纹理模型与均值漂移的改进分割算法。首先,对原始影像进行同等组滤波和颜色量化,得到颜色-纹理模型;其次,利用均值漂移算法对滤波影像进行初始分割,得到同质性较好的初始分割区域;最后,将颜色-纹理模型及初始分割对象轮廓信息应用于区域合并过程中,结合形状特征增强分割对象的紧密性。该算法充分结合了图像的颜色、纹理特征,通过对不同类别的遥感影像的分割实验进行分析,结果表明分割效率和分割质量均得到较大提升,且具有较好的适用性、可靠性及精确性,对遥感影像中纹理信息丰富的植被、密集建筑区等具有较好的分割效果。

Given the drawbacks that conventional mean shift algorithm does not consider texture feature while clustering in feature space, a new algorithm based on mean shift and color-texture pattern was proposed in this paper. It first got the color-texture pattern after peer group filtering and color quantization; then, the mean shift clustering was conducted to get homogeneous regions as the initial segmenta- tion results; eventually, the color-texture pattern and initial segmentation object contours were applied in the process of region merging, combining shape feature which got tight segments. New algorithm fully combined color and texture features, the segmentation results got by different classes of remote sensing images demonstrated that the method was effective and had better applicability, reliability and accuracy. New algorithm got satisfactory results in rich texture regions such as vegetation and densely build areas.