ConvolutionKernel Class

Initializes a new instance of the ConvolutionKernel class.

Gets the emboss matrix [-2, -1, 0; -1, 1, 1; 0, 1, 2], which lights the image from the bottom right. Its weights add up to 1, so flat areas keep their color.

Gets the number of rows.

Gets a value indicating whether the matrix is applied in two passes, a row then a column. It is the case of a matrix created by ConvolutionKernel.CreateSeparable, and of a matrix given weight by weight whose every weight is exactly, in double precision, the product of a weight of its largest weight's column, divided by that weight, and a weight of its row (a box, a binomial or a Sobel matrix): finding the factors costs one pass over the weights.

Gets the Laplacian of the four nearest neighbors, the sum of the second derivatives [1, -2, 1] of both directions: [0, 1, 0; 1, -4, 1; 0, 1, 0], as cv2.Laplacian with ksize = 1. Its weights add up to zero, and it is negative on the bright side of an edge.

Gets the 5x5 Laplacian, the sum of the second derivatives [1, 0, -2, 0, 1] of each direction smoothed by [1, 4, 6, 4, 1] across it (the second derivatives of the 5x5 Sobel operator), as cv2.Laplacian with ksize = 5: [2, 4, 4, 4, 2; 4, 0, -8, 0, 4; 4, -8, -24, -8, 4; 4, 0, -8, 0, 4; 2, 4, 4, 4, 2]. Its weights add up to zero.

Gets the 5x5 Laplacian of Gaussian, the Laplacian of a slightly blurred image, which responds less to noise: [0, 0, 1, 0, 0; 0, 1, 2, 1, 0; 1, 2, -16, 2, 1; 0, 1, 2, 1, 0; 0, 0, 1, 0, 0]. Its weights add up to zero. It is the classic 5x5 approximation with the sign of the Laplacian (negative in the middle, like ConvolutionKernel.Laplacian3x3); the "Mexican hat" form of that approximation is its negation.

Gets the horizontal Prewitt operator, the derivative [-1, 0, 1] of the rows averaged across them: [-1, 0, 1; -1, 0, 1; -1, 0, 1].

Gets the vertical Prewitt operator, ConvolutionKernel.PrewittX transposed: [-1, -1, -1; 0, 0, 0; 1, 1, 1].

Gets the horizontal Scharr operator, a Sobel operator whose smoothing [3, 10, 3] makes the response closer to the same for every direction of the edge: [-3, 0, 3; -10, 0, 10; -3, 0, 3], as cv2.Scharr with dx = 1 and dy = 0.

Gets the vertical Scharr operator, ConvolutionKernel.ScharrX transposed: [-3, -10, -3; 0, 0, 0; 3, 10, 3].

Gets the sharpen matrix [0, -1, 0; -1, 5, -1; 0, -1, 0]: the pixel minus the Laplacian of its four nearest neighbors. Its weights add up to 1, so flat areas keep their color.

Gets the horizontal Sobel operator, the derivative [-1, 0, 1] of the rows smoothed by [1, 2, 1] across them: [-1, 0, 1; -2, 0, 2; -1, 0, 1]. It responds to vertical edges, positively where the right side is brighter, as cv2.Sobel with dx = 1 and dy = 0.

Gets the vertical Sobel operator, ConvolutionKernel.SobelX transposed: [-1, -2, -1; 0, 0, 0; 1, 2, 1]. It responds to horizontal edges, positively where the bottom side is brighter.

Gets the number of columns.

Gets a weight.

Creates the matrix of a motion blur: a straight segment of length pixels through the center, at degrees from the horizontal, as the trail of a camera or a subject that moved while the shutter was open. Apply it with ImageProcessingExtensions.Convolve.

Creates a separable matrix: the weight at column x and row y is columnWeights[y] * rowWeights[x]. It is applied in two passes (every row filtered by rowWeights, then the filtered rows combined by columnWeights), so the time grows with the sum of the two lengths instead of their product.