Paper-Ranking Saliency
2017-10-03 15:46:38 2 举报
登录查看完整内容
pipline of Saliency detection by manifold ranking
作者其他创作
大纲/内容
Stage 2: foreground quary
- 阈值大于均值的点作为前景种子点，然后传播标签
- 标签传播的方法与前面相同：W*Label- assign saliency value to each pixel
design graph model
- compute feature for each superpixels
+ operation1(params):returnType- operation2(params)- operation3()
set parameters
- superpixel number: 200- edge weight: 10- maniford cost function weight coefficient: 0.99
Stage 1: background quary
- 分别将四个边界作为种子点，进行标签传播，然后通过乘积合并。- 将得到的显著性值安排到每个pixel。（通过元胞数组存储每个超像素的标签值）
generate superpixels
调用SLICSuperpixelSegmentation
生成- superpixels Image- acctual number of superpixel
0 条评论
回复 删除
下一页
