Difference between revisions of "Example Convolution"

From BoofCV
Jump to navigationJump to search
m
m
 
Line 2: Line 2:


Example Code:
Example Code:
* [https://github.com/lessthanoptimal/BoofCV/blob/v0.39/examples/src/main/java/boofcv/examples/imageprocessing/ExampleConvolution.java ExampleConvolution.java]
* [https://github.com/lessthanoptimal/BoofCV/blob/v0.40/examples/src/main/java/boofcv/examples/imageprocessing/ExampleConvolution.java ExampleConvolution.java]


Concepts:
Concepts:
Line 24: Line 24:


public static void main( String[] args ) {
public static void main( String[] args ) {
BufferedImage image = UtilImageIO.loadImage(UtilIO.pathExample("sunflowers.jpg"));
BufferedImage image = UtilImageIO.loadImageNotNull(UtilIO.pathExample("sunflowers.jpg"));


GrayU8 gray = ConvertBufferedImage.convertFromSingle(image, null, GrayU8.class);
GrayU8 gray = ConvertBufferedImage.convertFromSingle(image, null, GrayU8.class);

Latest revision as of 15:59, 17 January 2022

Example of how to convolve 1D and 2D convolution kernels across an image. Besides providing the kernel, how the border is handled needs to be specified. A normalized kernel will renormalize the

Example Code:

Concepts:

  • Convolution
  • Spacial filtering

Relevant Examples:

Example Code

/**
 * Several examples demonstrating convolution.
 *
 * @author Peter Abeles
 */
public class ExampleConvolution {

	private static final ListDisplayPanel panel = new ListDisplayPanel();

	public static void main( String[] args ) {
		BufferedImage image = UtilImageIO.loadImageNotNull(UtilIO.pathExample("sunflowers.jpg"));

		GrayU8 gray = ConvertBufferedImage.convertFromSingle(image, null, GrayU8.class);

		convolve1D(gray);
		convolve2D(gray);
		normalize2D(gray);

		ShowImages.showWindow(panel, "Convolution Examples", true);
	}

	/**
	 * Convolves a 1D kernel horizontally and vertically
	 */
	private static void convolve1D( GrayU8 gray ) {
		var kernel = new Kernel1D_S32(2);
		kernel.offset = 1; // specify the kernel's origin
		kernel.data[0] = 1;
		kernel.data[1] = -1;

		var output = new GrayS16(gray.width, gray.height);

		GConvolveImageOps.horizontal(kernel, gray, output, BorderType.EXTENDED);
		panel.addImage(VisualizeImageData.standard(output, null), "1D Horizontal");

		GConvolveImageOps.vertical(kernel, gray, output, BorderType.EXTENDED);
		panel.addImage(VisualizeImageData.standard(output, null), "1D Vertical");
	}

	/**
	 * Convolves a 2D kernel
	 */
	private static void convolve2D( GrayU8 gray ) {
		// By default 2D kernels will be centered around width/2
		var kernel = new Kernel2D_S32(3);
		kernel.set(1, 0, 2);
		kernel.set(2, 1, 2);
		kernel.set(0, 1, -2);
		kernel.set(1, 2, -2);

		// Output needs to handle the increased domain after convolution. Can't be 8bit
		var output = new GrayS16(gray.width, gray.height);

		GConvolveImageOps.convolve(kernel, gray, output, BorderType.EXTENDED);
		panel.addImage(VisualizeImageData.standard(output, null), "2D Kernel");
	}

	/**
	 * Convolves a 2D normalized kernel. This kernel is divided by its sum after computation.
	 */
	private static void normalize2D( GrayU8 gray ) {
		// Create a Gaussian kernel with radius of 3
		Kernel2D_S32 kernel = FactoryKernelGaussian.gaussian2D(GrayU8.class, -1, 3);
		// Note that there is a more efficient way to compute this convolution since it is a separable kernel
		// just use BlurImageOps instead.

		// Since it's normalized it can be saved inside an 8bit image
		var output = new GrayU8(gray.width, gray.height);

		GConvolveImageOps.convolveNormalized(kernel, gray, output);
		panel.addImage(VisualizeImageData.standard(output, null), "2D Normalized Kernel");
	}
}