COMZYH, computer Major in Reading 126 people agree to see the answer to the tree, with Mathematica did the next experiment, the effect is not so good.
First enter the image
Do discrete Fourier transforms:
data = imagedata[colorseparate[image][[1]];
{nrow, Ncol} = dimensions[data]
fw = Fourier[data];
LISTDENSITYPLOT[ABS@FW]
The result is this:
Then we take a low pass, first make a mask, here only take the entire picture 1/16 parts of the low frequency:
low = 0.25;
Mask = table[
if[abs[(Row-(NROW/2))/nrow] > Low &&
abs[(Col-(NCOL/2))/ncol] > Low, 1, 0], { Row, 1, nrow}, {col,
1, ncol}];
Masks (mask) are as follows:
As long as the white part does not have the blue part
Well, for this low-pass image do
Inverse Fourier transformand draw it out:
image[abs@ Inversefourier[fw*mask]]
Original
Low = 0.25
Low = 0.3
Low = 0.4
Using a two-Gaussian to filter the image after the DfT, suppress the low frequency, the result is as follows
Mask2 = imagedata[
imageresize[
image[rotateleft[
transpose[
rotateleft[gaussianmatrix[100] *14341.83641834123] [], []],
{ncol, nrow}];
It can be seen that using the Fourier method to do low-pass, the effect is not so good, of course, it may be my posture is not right.
If you want to try it yourself, the source code (MATHEMATICA) is as follows:
data = imagedata[colorseparate[image][[1]];
{nrow, Ncol} = dimensions[data]
fw = Fourier[data];
LISTDENSITYPLOT[ABS@FW] Low
= 0.3;
Mask = table[
if[abs[(Row-(NROW/2))/nrow] > Low &&
abs[(Col-(NCOL/2))/ncol] > Low, 1, 0], { Row, 1, nrow}, {col,
1, ncol}];
Listdensityplot[mask]
Listdensityplot[abs*mask]
image[abs@ Inversefourier[fw*mask]]
The code references the following answers and is interested to see for yourself:
Calculate the 2D Fourier transform of an Image
How to use 2D Fourier analysis to clean the noise in an image
Do a little work, welcome to correct me. Edited on 03:29 29 reviews thanks for sharing collection • No help • report • Author retention Rights 79 agree against, will not show your name Xiao Ji, PhD student in astronomy and Astrophys ... 79 people agree with the general FFT repair pictures are some of the rules of the pattern and special pollution, like the picture of the tree but I am curious how to repair. For example, the picture below shows a very regular stripe on the photo. Then there will be a very regular point on the FFT spectrum graph. These points are the correspondence of stripes in the frequency domain space.
If you erase these points and do an FFT inverse transformation, you can restore the original image well. But, inevitably, the image gets a little blurry ~
In general, the high frequency leaves the image details. The low frequency leaves the image as a whole. Filtering forever only causes the image to lose more information than to add details.
However, for the following picture, the frequency graph is too strange, the noise is not regular. The traditional FFT has no way of eliminating noise at all.
After many attempts and @Comzyh consistent. It is impossible to eliminate noise by filtering high-frequency signals.
Compared to the original answer has two images of the FFT value, can be seen, so-called filter results of the picture instead of a lot of high-frequency signal information (black stripes). So I guess the second picture below is the original, and the first one is the picture after the noise. This noise is not a good deal of the FFT, as described earlier, the FFT is good at eliminating the regularity of pollution and noise.
The female owner of the tree is the Korean actress son Ye Jin Jean, who starred in the summer aroma, and is clearly a photograph taken by a large-aperture SLR or HD camera. It is too difficult to say that the FFT can fix the noise to such details.
Finally, the software used is ImageJ, but now the Photoshop plugin also supports FFT.
PS: In cosmology, the discrete Fourier transform is an important application in the numerical simulation method, and it is the core algorithm of the particle Mesh method. The core idea is to map irregular particles to normal grids, and to quickly calculate the force and gravitational potential between particles by Fourier variation, which can greatly compress the N-body particle operations. Edited on 13:56 13 reviews thanks for sharing collection • No help • report • Author retention Rights 71 agree against, will not show your name Su Xiao, composed of carbohydrates and a small amount of metal impurities ♿ 71 people agree I'm sorry, please allow me to be unkind once: @ Trees but this top answer is problematic, at least with a picture of false.
In fact, people who have done image processing should know that noise feature extraction must not be simple can be calculated, noise reduction after the image is generally blurred, it is impossible to noise without a picture is bigger and clearer than the original. Also, the time domain to the frequency domain and then to the time domain conversion, the middle has done the data processing, inevitably will occur the phenomenon of distortion, with the diagram also did not reflect, so I decided this is a water answer.
I'm going to have time to comb some of the applications of FFT: including continuous signal sampling, image processing feature analysis, circuit filtering and harmonic analysis.
* Updated on the next day
@Comzyh classmates are very serious. There is suspicion on the hands to test, this is the technical person should be rigorous attitude, everyone interested please go here to see-
http://zhihu.com/question/20460630/answer/105852333
