纹理图像的多尺度分割

纹理图像分割最常用的方法就是基于特征的纹理分割,即首先提取出图像的纹理特征,在利用提取出的纹理特征来进行特征划分.通过对纹理特征进行划分来实现纹理图像分割的过程所面对的主要困难可以概括为:效率和效果.纹理图像分割算法大多具有较高的时间复杂度,这一方面是因为纹理特征提取比较费时,另一方面较高的特征维数导致特征划分过程的计算量通常比较大.本文提出基于图像四叉树的多尺度分割算法来实现实时图像的粗分割,实验表明此算法可以在保持分割精度的前提下大大降低时间复杂度.

Multi-scale Segmentation of Texture Image

The basic method of segmentation of texture image is segmentation based on extracted features.At first,texture features were extracted based texture elements;then,we use these features to locate pixel.The difficulty of segmentation based on features are effect and efficiency.The segmentation method of texture image has more time complexity: one is that the method of feature extraction is time-consuming;another is that high feature dimension results the computation of feature division is huge.To reduce the time complexity,a new quadtree-based multiscale segmentation algorithm were proposed in this paper.The experimental results demonstrate that proposed approach can provide more effective and more efficient segmentations.