Image processing using the frequency domain - duplicate image

2 次查看（过去 30 天）

显示更早的评论

Despairy 2013-1-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/59269-image-processing-using-the-frequency-domain-duplicate-image

评论： gromit park 2016-10-13

采纳的回答： Matt J

I need to understand the way this picture can be fixed

need to get only one instance of the dog.

This is what we have learned :

low pass filter
high pass filter
homomorphic filter

is there any way using the above to fix the image? I've included the fft image so maybe you can point out the thing I should be seeing that makes the image blurry as it is.

asking for theoretical help mostly , thanks

2 个评论
显示无隐藏无

Image Analyst 2013-1-19

Avoid the problem in the first place. Lay off the caffeine so you can keep the camera still, or train Sean's dog to be more still. ;-)

gromit park 2016-10-13

Th right FFT is the result of left image?

请先登录，再进行评论。

请先登录，再回答此问题。

采纳的回答

Matt J 2013-1-19

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/59269-image-processing-using-the-frequency-domain-duplicate-image#answer_71717

编辑：Matt J 2013-1-19

在 MATLAB Online 中打开

If F(k) is the spectrum of the dog, then F(k)*exp(-2*pi*j*dot(t,k)) is the spectrum of the dog translated by t and the spectrum of the blurry dog is their superposition

F(k)*(1+exp(-2*pi*j*dot(t,k)))

I suspect you're being asked to filter out the (1+exp(-2*pi*j*dot(t,k))). You need to know t in order to do so, but t dictates the orientation of the sinusoidal pattern of lines you're seeing in the background in the spectrum you've shown. Some sort of edge/line detection should do it.

10 个评论
显示 8更早的评论隐藏 8更早的评论

Image Analyst 2013-1-20

Have you ever looked at the spectrum without fftshift()? If so, you'll notice the "origin" is at the 4 corners so your filter will have to go out only until the middle of the image and you'll have to flip things around as you process the other 3 quadrants so that their origins are in the same place, then you'll have to unflip then to put their origins back. It might be easier to fftshift your filter, but if you don't you'll have to extract quadrants, call things like flipud(), fliplr() or transpose to get the origin where it needs to be.

Despairy 2013-1-21