摘要： 针对当前像素级别的图像语义分割算法难以利用全局形状特征,导致分割对象轮廓模糊,造成错误识别的 问题,提出一种区域级别的基于纹理基元块识别与合并的图像语义分割算法。该算法采用纹理基元等特征,考虑 到相邻像素点间的相互关系,保留物体间的棱角和边缘信息,分割出轮廓清晰的对象。在MSRC 图片库上进行实 验,结果表明,该算法能对多种语义对象进行分割和识别,具有运行速度快、识别率高和分割效果好等优点。

关键词: 纹理基元, 特征字典, k-d 树最近邻搜索, k-means 算法, 识别, 合并, 语义分割

Abstract: Aiming at the problem that the current image semantic segmentation algorithm at pixel level is difficult to use global shape features,leading the fuzzy contour of object and some wrong recognitions. This paper presents a new regional level image semantic segmentation algorithm based on texture element block recognition and merging. This algorithm uses the texture element feature to segment objects with a clear outline,which fully considers the relationship between adjacent pixels and keeps corners and edge information between objects. Experiments conducted on the MSRC database show that this method can segment and recognize a variety of semantic. Besides,it has the advantages of high efficiency,high recognition rate and good segmentation effect.

Key words: texture element, feature dictionary, k-d tree nearest neighbor search, k-means algorithm, recognition, merging, semantic segmentaion

中图分类号: 

• TP391. 41