Paper reading :Hierarchical Saliency Detection
来源:互联网 发布:领奖励的软件 编辑:程序博客网 时间:2024/05/21 12:48
Overview of hierachical framework-2013
Image Layer Extraction
—————coarse representation of the input with different degrees of details
1. define scale of region R
2. Efficient Algorithm to Compute Region Scale
避免穷举法计算带来的时间代价,采用box filter with t x t kernel to the segment and cheking if all pixel values inside the segment are changed during filtering.
3. Layer Generation
First, over segmentation : by the watershed like method. For each segmented region, we compute a scale value
Single-Layer Saliency Cues
———————saliency cues are applied to find important pixels from the perspectives of color, position and size
1. Local contrast
ci and cj are colors of regions Ri and Rj respectively. ( Lab color space)
w(Rj) : the number of pixels in Rj.
φ(i, j): is set to exp{−D(Ri, Rj)/σ2}, Parameter σ2 is set to the product of (0.2)2 and the particular scale threshold for the current layer.
2. Location heuristic——pixels close to a natural image center could be salient in many cases
where {x0, x1 · · · } is the set of pixel coordinates in region Ri;
λ is set to 9 in general (控制);
xc is the coordinate of the image center
local contrast and location cues normalized to range [0,1)
3. Initial saliency value
initial saliency cue maps in three layers
Hierarchical Inference——in 2013
多层融合时不采用简单的加权,而是用分层推导的方法得出能量方程,最后采用置信传播算法最小化能量方程。
For a node corresponding to region i in layer Ll, we define a saliency variable sli.
minimize the following energy function:
Data term ED(sli) is to gather separate saliency confidence.
- βl controls the layer confidence
Hierarchy term ES(sli,sl+1j) enforce consistency between corresponding regions in different layers.
(notice: Rli and Rl+1j are corresponding in two layers, we have Rli ⊆ Rl+1j)
Finally, minimize Eq.(5) by belief propagation.
Overview of hierachical framework-2013
对比 :
13年:对于L层i节点显著性值考虑与初始显著性值差距最小化,同时与L+1层对应父节点显著性值差异最小化;
15年:在13年的基础上,加上一条规则:对于L层i节点考虑与L层相邻的所有节点 j 依据相似性缩小差距(相似度大,差距越小),用来平滑相似的相邻区间的显著性值
效果上的前后对比:
Hierarchical Inference——in 2015
- new energy function :
- Data term ED(sli) and Hierarchy term EH(sli,sl+1j) 几乎与2013年的一样,不同之处在于添加了权重参数:
βi is fixed as {0.5, 4, 2} for i = {1, 2, 3};
- Local Consistency term EC(sli,sl+1j) which enforces intra-layer smoothness. It is used to make saliency assignment smooth between adjacent similar regions.
Notation A(Rki):Rki的邻域集合。which means Rkj∈ A(Rki) if Rkj is the adjacent region of Rki
γk determines the strength of consistency for each layer.
(in the LUV color space)σc is the normalization parameter.
Finally, optimaization.(有点儿复杂,以后再写,而且文中有很多的参数还没太弄明白,以后再更新)
- Paper reading :Hierarchical Saliency Detection
- Saliency Detection几篇paper
- #Paper Reading# Incremental Algorithms for Hierarchical Classification
- Reading Paper for Pedestrian detection 3.22-3.27
- Paper Reading - Mobile Security Survey & DS Failure Detection
- Paper reading:Global Contrast based Salient Region Detection
- Paper Reading
- Paper Reading
- Paper-Reading
- [Paper Reading]开始写Paper Reading Report
- Saliency Detection on Light Field
- 显著性检测(Saliency Detection)
- CVG Talk - Saliency based detection
- Saliency Detection with Caffe Framework
- After paper reading.......
- Deep Learning paper reading
- OSDI 2014 paper reading
- 写在Paper Reading之前
- ArcGIS Engine 创建索引(属性索引)
- Linux MTD系统剖析
- oracle 数据库升级
- Installing ADB on MAC OS X
- 机器学习读书笔记(一):绪论
- Paper reading :Hierarchical Saliency Detection
- Android时间对话框
- Android:改变 Toolbar 的文字和溢出图标颜色
- 数据库优化实例以及优化的小技巧
- Linux五个查找命令
- oracle 表空间、用户等相关基本操作
- iOS中CocoaPods继承使用Charts总结
- 在主方法中创建String对象,实现使用substring()方法对字符串进行截取
- java调用python方法