ImageCorrelate

gives the correlation of image with kernel ker.

computes a generalized correlation in which the function f is used in place of Dot.

Details and Options

ImageCorrelate performs the correlation operation on an image. It is a spatial filtering operation commonly used to find matches between images.
The correlation kernel ker is given as a two-dimensional numerical matrix, a one-channel image or a multichannel image with the same number of channels as image.
ImageCorrelate works with 2D and 3D images, operating separately on each channel.
ImageCorrelate[image,ker] by default gives an image of the same dimensions as image.
ImageCorrelate takes a Padding option. The default setting is Padding"Fixed".
With a setting PaddingNone, ImageCorrelate[image,ker] normally gives an image smaller than image.
ImageCorrelate gives an image of a real type.
In ImageCorrelate[image,ker,f], typical settings for the correlation operator f include:

	Dot	standard correlation
	ManhattanDistance	Manhattan or "city block" distance
	EuclideanDistance	Euclidean distance
	SquaredEuclideanDistance	squared Euclidean distance
	NormalizedSquaredEuclideanDistance	normalized squared Euclidean distance
	CosineDistance	angular cosine distance
	CorrelationDistance	correlation coefficient distance

ImageCorrelate takes a PerformanceGoal option to specify what aspect of performance to try to optimize. The default setting is PerformanceGoal"Speed".

Examples

open allclose all

Basic Examples (2)

Blur an image by correlating it with a Gaussian matrix:

Find occurrences of a template in an image:

Scope (5)

Data (3)

Correlate a binary image with a disk-shaped kernel:

Correlate a grayscale image:

Correlate a 3D image:

Parameters (2)

Use a Laplacian kernel:

Use a template:

The default correlation operator f is Dot:

Use a different correlation operator:

Generalizations & Extensions (2)

Nonrectangular templates are better matched when used with a mask:

The correlation obtained using a mask is more robust to background changes:

Result of the template matching without using the mask:

Match rotated occurrences of a template by matching the template rotated using a range of angles:

Visualize the best matches:

Result of the template matching without taking into account multiple orientations:

Options (3)

Padding (3)

By default, a "Fixed" padding is used:

Specify a custom padding:

Padding->None normally returns an image smaller than the input image:

Applications (2)

Isolate regions of interest in an image:

A basic face detector that uses the normalized correlation with an image of a pair of eyes:

Properties & Relations (1)

ImageCorrelate gives an image of a real type:

Introduced in 2008

(7.0)

Updated in 2010

(8.0)

2012

(9.0)

Top