一种局部纹理特征在区域生长中的应用*

为克服现有算法不能自动选择种子且没有很好地利用纹理信息的局限性，提出了一种基于局部纹理特征的区域生长算法。算法分为两个阶段，即颜色量化和区域生长。先利用Mean Shift算法对颜色聚类，对图像预分割，用该局部量化纹理算法提取量化结果的纹理特征，由该纹理特征自动选择种子并分级合并区域。与经典的JSEG算法相比，该算法能够得到相似的分割结果，且计算复杂度低。

Application of one local texture characteristic to region growing

In order to overcome these shortings,such as the thraditional algorithms could not select seeds automatically and apply texture information commendably,this paper proposed a novel region growing algorithm based on local texture characteristics.This method consisted of two independent steps:color quantization and region growing.In the first step,colors in the image were quantized to several representative classes appling Mean Shift method,thus forming pre-segmentation result.Appling the local quantized texture algorithm to pre-segmentation result to get a texture characteristic,then seeds could be automatically selected and a region growing method was then employed to segment the image based on this texture characteristic.Compared with the classical algorithm JSEG,experiments show high similar segmentation result and lower computational complexity.