Difference between revisions of "Example Wavelet Noise Removal"

From BoofCV
Jump to navigationJump to search
(Fixed type-o)
m
(11 intermediate revisions by the same user not shown)
Line 1: Line 1:
= Image Noise Removal with Wavelets =
<center>
<center>
<gallery widths=300px heights=300px>
<gallery widths=300px heights=300px>
Image:Example_lena_denoise_noisy.jpg | Image with Gaussian noise added to it.
File:Kodim17_noisy.jpg | Image with Gaussian noise added to it.
Image:Example_lena_denoise_denoised.jpg | After a Bayes Shrink.
File:Kodim17_denoised.jpg | After a Bayes Shrink.
</gallery>
</gallery>
</center>
</center>
Line 10: Line 8:
Image noise removal is the process of attempting to under the corruption caused by noise.  Often this noise is modeled as Gaussian noise being added to each pixel independently.  Several different wavelet algorithms have been proposed for removing noise from image.  These work by determining the statistical properties of the wavelet coefficients and removing the outliers.   
Image noise removal is the process of attempting to under the corruption caused by noise.  Often this noise is modeled as Gaussian noise being added to each pixel independently.  Several different wavelet algorithms have been proposed for removing noise from image.  These work by determining the statistical properties of the wavelet coefficients and removing the outliers.   


Except in highly constrained situations when strong assumptions can be made, it is never possible to recover the original image perfectly.  Wavelet based methods do have some advantages over other commonly used techniques.  For example, they tend to preserve edge features better than applying Gaussian blur.  To see a comparison between some of these techniques, click on the applet below.
Except in highly constrained situations when strong assumptions can be made, it is never possible to recover the original image perfectly.  Wavelet based methods do have some advantages over other commonly used techniques.  For example, they tend to preserve edge features better than applying Gaussian blur.  To see a comparison between some of these techniques, open the demonstration included with the project source code.


In the example below a simply interface is used to instantiate a wavelet based image denoiser.  This filter is then used to remove the noise and create a smoother image that appears less noisy.
In the example below a simply interface is used to instantiate a wavelet based image denoiser.  This filter is then used to remove the noise and create a smoother image that appears less noisy.


Example Code:
Example Code:
* [https://github.com/lessthanoptimal/BoofCV/blob/master/examples/src/boofcv/examples/ExampleWaveletDenoise.java ExampleWaveletDenoise.java]
* [https://github.com/lessthanoptimal/BoofCV/blob/v0.37/examples/src/main/java/boofcv/examples/enhance/ExampleWaveletDenoise.java ExampleWaveletDenoise.java]


Concepts:
Concepts:
* Noise Removal
* Noise Removal
* Wavelets
* Wavelets
Relevant Applets:
* [[Applet_Denoise| Image Denoising]]


= Example Code =
= Example Code =
Line 34: Line 29:
public class ExampleWaveletDenoise {
public class ExampleWaveletDenoise {


public static void main( String args[] ) {
public static void main( String[] args ) {


// load the input image, declare data structures, create a noisy image
// load the input image, declare data structures, create a noisy image
Random rand = new Random(234);
Random rand = new Random(234);
ImageFloat32 input = UtilImageIO.loadImage("../data/evaluation/standard/lena512.bmp",ImageFloat32.class);
GrayF32 input = UtilImageIO.loadImage(UtilIO.pathExample("standard/kodim17.jpg"),GrayF32.class);


ImageFloat32 noisy = input.clone();
GrayF32 noisy = input.clone();
GeneralizedImageOps.addGaussian(noisy,rand,20,0,255);
GImageMiscOps.addGaussian(noisy, rand, 20, 0, 255);
ImageFloat32 denoised = new ImageFloat32(input.width,input.height);
GrayF32 denoised = noisy.createSameShape();


// How many levels in wavelet transform
// How many levels in wavelet transform
int numLevels = 4;
int numLevels = 4;
// Create the noise removal algorithm
// Create the noise removal algorithm
WaveletDenoiseFilter<ImageFloat32> denoiser = FactoryImageDenoise.waveletBayes(ImageFloat32.class,numLevels);
WaveletDenoiseFilter<GrayF32> denoiser =
FactoryImageDenoise.waveletBayes(GrayF32.class,numLevels,0,255);


// remove noise from the image
// remove noise from the image
Line 58: Line 54:
gui.addImage(ConvertBufferedImage.convertTo(denoised,null),"Denoised");
gui.addImage(ConvertBufferedImage.convertTo(denoised,null),"Denoised");


ShowImages.showWindow(gui,"Wavelet Noise Removal Example");
ShowImages.showWindow(gui,"Wavelet Noise Removal Example",true);
}
}
}
}
</syntaxhighlight>
</syntaxhighlight>

Revision as of 12:53, 21 December 2020

Image noise removal is the process of attempting to under the corruption caused by noise. Often this noise is modeled as Gaussian noise being added to each pixel independently. Several different wavelet algorithms have been proposed for removing noise from image. These work by determining the statistical properties of the wavelet coefficients and removing the outliers.

Except in highly constrained situations when strong assumptions can be made, it is never possible to recover the original image perfectly. Wavelet based methods do have some advantages over other commonly used techniques. For example, they tend to preserve edge features better than applying Gaussian blur. To see a comparison between some of these techniques, open the demonstration included with the project source code.

In the example below a simply interface is used to instantiate a wavelet based image denoiser. This filter is then used to remove the noise and create a smoother image that appears less noisy.

Example Code:

Concepts:

  • Noise Removal
  • Wavelets

Example Code

/**
 * Example of how to "remove" noise from images using wavelet based algorithms.  A simplified interface is used
 * which hides most of the complexity.  Wavelet image processing is still under development and only floating point
 * images are currently supported.  Which is why the image  type is hard coded.
 */
public class ExampleWaveletDenoise {

	public static void main( String[] args ) {

		// load the input image, declare data structures, create a noisy image
		Random rand = new Random(234);
		GrayF32 input = UtilImageIO.loadImage(UtilIO.pathExample("standard/kodim17.jpg"),GrayF32.class);

		GrayF32 noisy = input.clone();
		GImageMiscOps.addGaussian(noisy, rand, 20, 0, 255);
		GrayF32 denoised = noisy.createSameShape();

		// How many levels in wavelet transform
		int numLevels = 4;
		// Create the noise removal algorithm
		WaveletDenoiseFilter<GrayF32> denoiser =
				FactoryImageDenoise.waveletBayes(GrayF32.class,numLevels,0,255);

		// remove noise from the image
		denoiser.process(noisy,denoised);

		// display the results
		ListDisplayPanel gui = new ListDisplayPanel();
		gui.addImage(ConvertBufferedImage.convertTo(input,null),"Input");
		gui.addImage(ConvertBufferedImage.convertTo(noisy,null),"Noisy");
		gui.addImage(ConvertBufferedImage.convertTo(denoised,null),"Denoised");

		ShowImages.showWindow(gui,"Wavelet Noise Removal Example",true);
	}
}