基于超像素的多视觉特征图像分割算法研究

Research on Multiple Visual Features Image Segmentation Algorithm Based on Superpixel

现有融合多种视觉特征的图像分割算法大多是将多种特征简单组成一组特征向量,如何针对不同图像自动调整多种特征之间的权重仍是一大挑战。基于此,提出一种基于区域的融合多视觉特征图像分割算法。首先,使用Meanshift算法对原始图像进行预分割,获得一组超像素区域。该操作既能保留图像局部的空间信息,又能降低时间复杂度;其次,分别提取每个超像素区域的颜色特征和纹理特征;之后,分别根据两个特征设计两个图像分割模型,并使用多目标进化算法对两个分割模型同时进行优化。将该算法与现有的特征融合分割算法从视觉与量化指标两方面进行对比,实验结果表明,该算法可很好地对多种特征进行融合,并取得了理想的分割效果。

The existing feature fusion image segmentation algorithms are usually to combine multiple visual features into a set of fea⁃ture vectors.How to automatically adjust the weights between multiple features for different images is still a challenge.Based on this,this paper proposes a region-based multi-visual feature fusion image segmentation algorithm.Firstly,the original image is pre-seg⁃mented using Meanshift algorithm to obtain a set of superpixel regions.This operation not only preserves the spatial information of the image,but also reduces the time complexity.Secondly,the color features and texture features of each superpixel region are extracted separately.Then,two image segmentation models are designed according to two features respectively,and the two segmentation models are simultaneously optimized using the multi-objective evolutionary algorithm.The proposed algorithm is compared with the existing feature fusion segmentation algorithms from the visual and quantitative indicators.The experimental results show that the proposed algo⁃rithm can integrate many features well and achieve satisfactory segmentation results.