Groups
Groups
Sign in
Groups
Groups
黃子嘉 - 線代離散研究室
Conversations
About
Send feedback
Help
線性代數,SVD奇異值分解在影像處理上的應用
1,246 views
Skip to first unread message
月戀星辰
unread,
Feb 22, 2013, 2:30:09 AM
2/22/13
Reply to author
Sign in to reply to author
Forward
Sign in to forward
Delete
You do not have permission to delete messages in this group
Copy link
Report message
Show original message
Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message
to zjh...@googlegroups.com
請問助教,我記得黃子嘉老師在最後一堂課有提到 SVD奇異值分解在影像處理上的應用,可是我不偷東西但剛好偷懶筆記沒抄,請問助教可以提醒一下嗎?
P.S. 我最近要重修影像處理
林立宇 (wynne助教)
unread,
Feb 22, 2013, 11:49:25 AM
2/22/13
Reply to author
Sign in to reply to author
Forward
Sign in to forward
Delete
You do not have permission to delete messages in this group
Copy link
Report message
Show original message
Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message
to zjh...@googlegroups.com
在影像處理, SVD主要是利用降低維度來將影像作壓縮
假設原先的影像矩陣 A: mxn 是 r 維
如果不想存那麼多資訊, 想將影像降到 k 維
則只需要砍掉後面 r - k 個singular value
也就是在
Σ 中
將最小的 r - k 個非零的singular value改成 0, 成為
Σ*
則我們可以證明, U
Σ*V^T就會是所有 k 維的mxn矩陣中最接近 A 的那一個
與A的error會最小,
因此比較不會讓影像失真
月戀星辰
unread,
Feb 22, 2013, 2:22:24 PM
2/22/13
Reply to author
Sign in to reply to author
Forward
Sign in to forward
Delete
You do not have permission to delete messages in this group
Copy link
Report message
Show original message
Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message
to zjh...@googlegroups.com
那為什麼降低維度就可以做壓縮呢?
把最小的 r-k 個 singular value 改為零後,為什麼就有壓縮的效果呢?
林立宇 (wynne助教)
unread,
Feb 23, 2013, 12:27:55 PM
2/23/13
Reply to author
Sign in to reply to author
Forward
Sign in to forward
Delete
You do not have permission to delete messages in this group
Copy link
Report message
Show original message
Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message
to zjh...@googlegroups.com
因為rank變小, 相對帶的資訊量就變少了呀
比方說把SVD看成老師在題庫班提過的那個形式, 寫成rank-1 matrix的和
i.e., A =
σ1u1v1^t +
σ2u2v2^t
+
... +
σrurvr^t
本來有那麼多項, 如果不想放那麼多資訊,
那我就取前面幾項相加就好了 (
要取幾項就看效果和儲存量要怎麼做取捨
)
定理告訴我這樣做出來的效果和 A 的誤差會最小
這種降維找近似於 A 的矩陣的方法又稱為low-rank approximation
月戀星辰
unread,
Feb 23, 2013, 9:54:32 PM
2/23/13
Reply to author
Sign in to reply to author
Forward
Sign in to forward
Delete
You do not have permission to delete messages in this group
Copy link
Report message
Show original message
Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message
to zjh...@googlegroups.com
原來如此! 感謝助教指點迷津
Reply all
Reply to author
Forward
0 new messages