Accelerating the pace of engineering and science

Documentation Center

• Trial Software

integralKernel class

Integral image filter

Description

This object describes box filters to use with integral images.

Construction

H = integralKernel(BBOX,WEIGHTS) defines a box filter, H, from an M-by-4 matrix of bounding boxes, BBOX, and an M-length vector, WEIGHTS, of corresponding weights. Each row of BBOX corresponds to a different region, which you can use to sum the pixels. The bounding boxes can overlap. The format of each row, [x y width height], contains the coordinates of the upper-left corner and size of each region. See Define a Vertical Derivative Filter for an example of how to specify a box filter.

Input Arguments

 BBOX M-by-4 matrix of bounding boxes. Each of the M number of 4-element row vectors [x y width height] define a bounding box. The first two elements represent the coordinates of the upper-left corner of the bounding box. The second two elements represent the width and height of the bounding box. WEIGHTS M length vector of weights corresponding to the bounding boxes.

Properties

 BoundingBoxes Bounding boxes which define the filter. The format is [x y width height]. Weights Vector containing a weight for each bounding box. Coefficients Conventional filter coefficients. Center Center coordinates of the filter. Size Filter size.

Methods

 transpose Transpose filter

Examples

expand all

Define a Vertical Derivative Filter

Specify a conventional filter with coefficients.

Specify a conventional filter with the following coefficients.

Use the format [x y width height] for the two bounding boxes.

 First region x = 1, y = 1, width = 4, height = 2, weight = 1 Second region x = 1, y = 3, width = 4, height = 2, weight = –1

Enter the specification as follows.

`boxH = integralKernel([1 1 4 2; 1 3 4 2], [1, -1]);`

Define an Average Filter

Define an 11-by-11 average filter.

`avgH = integralKernel([1 1 11 11], 1/11^2);`

Define a Filter Approximating a Second-Order Partial Derivative

Define filter which approximates second order partial derivative in XY direction. This filter can be used by the SURF feature detector.

```% The filter is zero-padded, using weight of zero, so that it is centered at x=5, y=5.
xydH = integralKernel([1,1,9,9; 2,2,3,3; 6,2,3,3; 2 6 3 3; 6 6 3 3], [0, 1, -1, -1, 1]);

% Visualize the filter
imshow(xydH.Coefficients, [], 'InitialMagnification', 'fit');

% Mark the filter center
hold on; plot(xydH.Center(1), xydH.Center(2), 'o');
impixelregion;```

Define a Filter to Approximate a Gaussian Second-order Partial Derivative

Define a filter to approximate a Gaussian second order partial derivative in Y direction.

```ydH = integralKernel([1,1,5,9;1,4,5,3], [1, -3]);

% Note that this same filter could have been defined as:
%   integralKernel([1,1,5,3;1,4,5,3;1,7,5,3], [1, -2, 1]);
% but it would be less efficient since it requires 3 bounding boxes.

ydH.Coefficients % visualize the filter```

References

Viola, Paul and Michael J. Jones, "Rapid Object Detection using a Boosted Cascade of Simple Features", Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2001. Volume: 1, pp. 511–518.