摘要： 提出一种符合人类视觉感知的图像对象分割方法，包括双尺度的区域分割和基于模型的对象提取。运用非线性尺度算子对图像进行大尺度平滑，结合颜色量化和视觉一致性的颜色聚类完成图像的粗分割。在原尺度上融合区域的纹理、颜色信息对分割区域进行区域合并，并利用对象模型完成图像对象的提取。实验结果表明，该算法的分割结果符合人类视觉感知特性，能够较好地完成图像对象分割。

关键词: 视觉感知, 对象分割, 双尺度空间, 区域合并准则

Abstract: Image object segmentation method in accord with human vision perception, which consists of double-scale region segmentation and object extraction based on object model, is proposed in this paper. An image is smoothed by nolinear scale operator in coarse-scale, color quantization and perceptual color clustering are exploited to form the initial segmented regions. The initial regions are merged based on the region distance in the original finest-scale. Object model is used to implement image object extraction. Experimental results demonstrate that the proposed algorithm gets better image object segmentation with favorable consistency to human vision perception.

Key words: vision perception, object segmentation, double-scale spaces, region merger criteria

中图分类号: 

• TP391