| 查看: 1738 | 回复: 4 | ||||
xiashaoyan新虫 (正式写手)
|
[求助]
图像的非线性扩散滤波倒是是怎么扩散的？ 已有1人参与
|
|
近一直看图像的非线性扩散，有一个问题一直困扰的自己，问题是：图像的非线性扩散滤波倒是是怎么扩散的？ 问题1：在各向异性扩散的PM模型中，通常边缘处的扩散系数比较小，而噪声处的扩散系数比较大，那么通过这样的设计，假设在噪声点(这个噪声点灰度值比较大)附近有边缘信息，那么经过扩散后，这个噪声会的灰度值降低了，那么降低的灰度值量是不是加到附近边缘信息的灰度上量，也就是，在扩散过程中，扩散是相邻两个点灰度值量的扩散吗？或者说，噪声点经过扩散后，扩散的量加到附近点上了吗？我看过一篇文献，说传统的PM扩散是一个绝热过程，只在相邻两个点之间进行扩散。问题2：如果正如我看到的文献所说的那样，一般的PM扩散是两个点的扩散，那么岂不是会改变一些边缘信息的灰度值吗？(因为相邻扩散的两个点是相互影响的) |
回复此楼» 猜你喜欢
化学专业申博
已经有4人回复
-
售SCI一区T0P文章，我:8.O.5.5.1.O.5.4,科目齐全,可+急
已经有3人回复
-
评审有感
已经有27人回复
-
论文撤稿了
已经有7人回复
-
售SCI一区T0P文章，我:8.O.5.5.1.O.5.4,科目齐全,可+急
已经有4人回复
-
售SCI一区T0P文章，我:8.O.5.5.1.O.5.4,科目齐全,可+急
已经有3人回复
-
售SCI一区T0P文章，我:8.O.5.5.1.O.5.4,科目齐全,可+急
已经有3人回复
-
售SCI一区T0P文章，我:8.O.5.5.1.O.5.4,科目齐全,可+急
已经有5人回复
-
售SCI一区T0P文章，我:8.O.5.5.1.O.5.4,科目齐全,可+急
已经有4人回复
-
河北省自然科学基金
已经有9人回复
» 本主题相关价值贴推荐，对您同样有帮助:
- 《MATLAB图像滤波去噪分析及其应用》源码共享
已经有83人回复
- 2015年南京航空航天大学国家自然科学基金资助清单
已经有14人回复
- 中国石油大学(华东)自然科学基金立项名单
已经有8人回复
- 2013西安电子科技大学NSFC全名单整理（136项）
已经有2人回复
- 西安电子科技大学清单
已经有17人回复
zaq123321
专家顾问 (著名写手)
-
专家经验: +343 - 数学EPI: 6
- 应助: 298 (大学生)
- 贵宾: 0.247
- 金币: 10791.3
- 散金: 280
- 红花: 29
- 帖子: 1223
- 在线: 539.9小时
- 虫号: 405284
- 注册: 2007-06-17
- 性别: MM
- 专业: 生物大分子结构与功能
- 管辖: 数学
【答案】应助回帖
★ ★ ★ ★ ★
感谢参与，应助指数 +1
xiashaoyan: 金币+5, ★有帮助 2015-09-27 23:08:19
感谢参与，应助指数 +1
xiashaoyan: 金币+5, ★有帮助 2015-09-27 23:08:19
|
It's an iteration. You need to decide how many steps you want to go. Thus to design a good stop criterion is not easy. Usually the iteration steps and step size should be reported in paper. How can you say just diffusion within two points. One image is 128x128 or 64x64 [ 发自手机版 http://muchong.com/3g ] |
2楼2015-09-27 00:39:56
feixiaolin
荣誉版主 (文坛精英)
-
专家经验: +518 - 应助: 942 (博后)
- 贵宾: 1.275
- 金币: 3880
- 散金: 58785
- 红花: 532
- 沙发: 11
- 帖子: 24215
- 在线: 2601.8小时
- 虫号: 2139575
- 注册: 2012-11-21
- 专业: 光学信息获取与处理
- 管辖: 数学
| http://c.360webcache.com/c?m=68a ... 83f67240&q=图像的非线性扩散滤波&u=http%3A%2F%2Fblog.csdn.net%2Fchenyusiyuan%2Farticle%2Fdetails%2F8710462%2F |
3楼2015-09-27 06:52:49
xiashaoyan
新虫 (正式写手)
- 应助: 10 (幼儿园)
- 金币: 2574.9
- 散金: 30
- 红花: 10
- 帖子: 498
- 在线: 184.1小时
- 虫号: 2507802
- 注册: 2013-06-14
- 专业: 计算机软件
|
I think the iteration steps and step size of every point in a image is same,so it is hard that this design is keep all point have a good a diffusion.And my quesion is that what is diffusion for PM model or AD diffusion?Because I read a paper said that:AD model considers relations only between neighborhoods, rather than among neighborhoods; therefore, the thermal diffusivity is independently calculated between a center node and each neighborhood.,but I don't undarstand this meaning,and now I think how diffusion process for a image,especially for a noise point and signal point.Finally, if you interest this quesions, you can add my qq：1403291906，or you can left you email/QQ,I very expect that I can interact with you,Thank you! |
4楼2015-09-27 23:23:38
zaq123321
专家顾问 (著名写手)
-
专家经验: +343 - 数学EPI: 6
- 应助: 298 (大学生)
- 贵宾: 0.247
- 金币: 10791.3
- 散金: 280
- 红花: 29
- 帖子: 1223
- 在线: 539.9小时
- 虫号: 405284
- 注册: 2007-06-17
- 性别: MM
- 专业: 生物大分子结构与功能
- 管辖: 数学
【答案】应助回帖
|
what is diffusion for PM model or AD diffusion? I think PM model is a diffusion model. What's AD diffusion? AD model considers relations only between neighborhoods, rather than among neighborhoods My guess is that: Just use 5 points discretization or 9 points discretization for a point, or 13 points? the thermal diffusivity is independently calculated between a center node and each neighborhood My guess is that: For the center node, after you calculated and get a new value, whether you want to use this new value to update next pixel value or not? If using the new calculated value, then it's not independent. If using the value before updated, it's independent. I think how diffusion process for a image,especially for a noise point and signal point. Diffusion process doesn't know what's signal, what's noise. You need to figure out what's the difference between signal and noise and get their characteristic and put them in the diffusion model. That's the idea. So for isotropic model, it smoothes equally for noise and signal. For anisotropic model, it treats edges etc as features or signals and tries to preserve them and do less smooth. For other parts treat as noises and do harsh smooth. |
5楼2015-09-28 00:24:17
10