基于多尺度分析的归一化割图像分割算法

不同尺度的图像拥有不同的特性,针对单一尺度图像进行分割容易出现过分割或欠分割的问题,本文提出一种基于多尺度分析的归一化割的图像分割方法,首先利用方向能量模型得到不同尺度子图像的边缘方向能量,然后根据干涉轮廓的思想建立各个子图像像素之间的相似度,形成多个不同尺度的权值矩阵,并归一化为一个权值矩阵,最后运用归一化割算法对图像进行分割。实验表明,本文方法在融合了多个尺度图像不同特性的同时,能很好地处理含有纹理区域和弱边缘的图像,在一定程度上避免了过分割或欠分割的问题。

An image segmentation algorithm based on multi-scale analytic normalized cut

Images of different scales have different characteristics,the over-segmentation or incomplete segmentation may happen when segmented based on the image of single scale.So,normalized cut algorithm based on multi-scale analysis is proposed.First,oriented energy model is proposed to obtain sub-image of different scales and the image edge energy.Second,based on the intervening contour,the affinity between the pixels of the sub-image and the weight matrix of different scales is established,then normalized into a weight matrix.Third,normalized cut is used to segment images.The experimental results show that the algorithm that combine different characteristics of each scale image,has a good segmentation of the images which contained textured regions and weak contours,solves the over-segmentation or incomplete segmentation problems to some extent as well.