Template Matching

Locate a template in an image

• Library:

• Computer Vision Toolbox / Analysis & Enhancement

• 

Description

The Template Matching block finds the best match of a template within an input image. The block computes match metric values by shifting a template over a region of interest or the entire image, and then finds the best match location.

The block outputs either the match metric values or the one-based (x,y) coordinates of the best template match. Optionally the lbock can output an N-by-N matrix of the match metric values centered around the best match location.

Ports

Input

I — Image

image

Input image to use with the template. The block does not pad the input data. Therefore, it can only compute values for the match metrics between the input image and the template, where the template is positioned such that it falls entirely on the input image. A set of all such positions of the template is termed as the valid region of the input image. The size of the valid region is the difference between the sizes of the input and template images plus one.

sizevalid=sizeinput – sizetemplate+1

(1)

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | Boolean | fixed point

T — Template

M-by-N matrix

Template, specified as an M-by-N matrix

The Template Matching block does not pad the input data. Therefore, it can only compute values for the match metrics between the input image and the template, where the template is positioned such that it falls entirely on the input image. A set of all such positions of the template is termed as the valid region of the input image. The size of the valid region is the difference between the sizes of the input and template images plus one.

sizevalid=sizeinput – sizetemplate+1

(2)

Data Types: single | double | uint8 | Boolean | fixed point

ROI — Region of interest

four-element vector

Region of interest vector in the format (x,y,width,height), where (x,y) are one-based coordinates for the upper-left corner of the region. The block outputs the best match location index relative to the top left corner of the input image.

Data Types: single | double | uint8 | Boolean | fixed point

Output

Metric — Match metric values

matrix

Matrix of match metric values. The matrix of the match metrics always implements single-step exhaustive window iteration. Therefore, the block computes the metric values at every pixel.

Dependencies

When you set the Output parameter to Metric matrix, the block outputs the valid image size.

Data Types: single | double | uint8 | Boolean | fixed point

Loc — Best match locations

two-element vector

Best template match located at the one-based (x,y) coordinates. When in the ROI processing mode, the block treats the image around the ROI as an extension of the ROI subregion. Therefore, it computes the best match locations true to the actual boundaries of the ROI. The block outputs the best match coordinates, relative to the top-left corner of the image. The one-based [x y] coordinates of the location correspond to the center of the template. The following table shows how the block outputs the center coordinates for odd and even templates:

Odd number of pixels in template	Even number of pixels in template

Data Types: uint32

NMetric — Metric values in neighborhood of best match

N-by-N matrix

Metric values in the neighborhood of the best match, returned as an N-by-N matrix. N, specified in the mask, must be an odd number. When you select Best match location to return the matrix of metrics in a neighborhood around the best match, an exhaustive loop computes all the metric values for the N-by-N neighborhood. This output is particularly useful for performing template matching with subpixel accuracy.

Data Types: single | double | uint8 | Boolean | fixed point

NValid — Valid neighborhood

off | on

Enable to track the valid neighborhood region. The neighborhood matrix of metric values is valid inside of the ROI. The block sets the NValid output as follows:

• 1 — The neighborhood containing the best match is completely inside the region of interest.

• 0 — The neighborhood containing the best match is completely or partially outside of the region of interest.

Dependencies

This port appears when you enable the Output NxN matrix of metric values around best match parameter.

Data Types: Boolean

ROIValid — Valid region of interest

off | on

Enable to track the valid region of interest. If the ROI lies partially outside the valid image, the block only processes the intersection of the ROI and the valid image. The block sets the ROIValid output as follows:

• 1 — The ROI lies completely inside the valid part of the input image.

• 0 — The ROI lies completely or partially outside of the valid part of the input image.

Dependencies

This port appears when you enable the Output flag indicating if ROI is valid parameter.

Data Types: Boolean

Parameters

Main Tab

Match metric — Match metric

Sum of absolute differences (default) | Sum of squared differences | Maximum absolute difference

Select match metric as one of:

• Sum of absolute differences (SAD)

• Sum of squared differences (SSD)

• Maximum absolute difference (MaxAD)

The block computes the match metric at each step of the iteration. Choose the match metric that best suits your application. The block calculates the global optimum for the best metric value. It uses the valid subregion of the input image intersected by the ROI, if provided.

Output — Match metric

Match metric (default) | Best match location

Select the type of output as one of:

• Match metric — Output the match metric matrix. This option adds the Metric output port to the block.

• Best match location — Output the [x,y] coordinates for the location of the best match. This option adds the Loc output port to the block. It also makes the Search method, Output NxN matrix of metric values around best match, and Enable ROI processing parameters available.

  The Output parameter on the Data Types pane appears when you set the Output parameter on the Main tab to Metric matrix or if you set it to Best match location and you enable the Output NxN matrix of metric values around best match parameter.

Search method — Search method

Exhaustive (default) | Three-step

Specify the search method as either Exhaustive or Three-step. The Exhaustive search method is computationally intensive because it searches at every pixel location of the image. However, this method provides a more precise result.

The Three-step search method is a fast search that uses a neighborhood approach versus a search at every pixel. The search starts with a step size equal to or slightly greater than half of the maximum search range and then employs the following steps:

1. The block compares nine search points in each step. There is a central point and eight search points located on the search area boundary.

2. The block decrements the step size by one, after each step, ending the search with a step size of one pixel.

3. At each new step, the block moves the search center to the best matching point resulting from the previous step. The number one blue circles in the figure below represent a search with a starting step size of three. The number two green circles represent the next search, with step size of two, centered around the best match found from the previous search. Finally, the number three orange circles represent the final search, with step size of one, centered around the previous best match.

Dependencies

This parameter appears when you set the Output parameter to Best match location.

Output NxN matrix of metric values around best match — Output N-by-N matrix

on (default) | off

Enable to add the NMetric and NValid ports.

Dependencies

This parameter appears when you set the Output parameter to Best match location.

N — Size of output matrix

3 (default) | integer

Size of output matrix. This value determines the size of the N-by-N output matrix centered around the best match location index. N must be an odd number.

Dependencies

This parameter appears when you enable the Output NxN matrix of metric values around best match parameter.

Enable ROI processing — Enable ROI processing

on (default) | off

Enable ROI processing.

Dependencies

• This parameter appears when you set the Output parameter to Best match location.

• When you enable this parameter, the block adds the ROI input port, and the Output flag indicating if ROI is valid parameter appears.

Output flag indicating if ROI is valid — Valid ROI

off (default) | on

Enable to add the ROIValid port.

Dependencies

This parameter appears when you enable the Enable ROI processing parameter.

Data Types Tab

For details on the fixed-point block parameters, see Specify Fixed-Point Attributes for Blocks .

Block Characteristics

Data Types	Boolean | double | fixed point | integer | single
Multidimensional Signals	no
Variable-Size Signals	yes

Algorithms

Match Metrics

The match metrics use a difference equation with general form:

$$d_p(x,y)=({\displaystyle \sum _{i=1}^n{\left|x_i-y_i\right|}^p{)}^{\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$p$}\right.}}$$

$$l_n^p$$ denotes the metric space $$(R^n,d_p)$$ for $$R^n$$ n > 1.

Sum of Absolute Differences (SAD)

The SAD metric is also known as the Taxicab or Manhattan Distance metric. It sums the absolute values of the differences between pixels in the original image and the corresponding pixels in the template image. This metric is the $$l^1$$ norm of the difference image. The lowest SAD score estimates the best position of template within the search image. The general SAD distance metric becomes:

$$d_1(I_j,T)={\displaystyle \sum _{i=1}^n\left|I_{i,j}-T_i\right|}$$

Sum of Squared Differences (SSD)

The SSD metric is also known as the Euclidean Distance metric. It sums the square of the absolute differences between pixels in the original image and the corresponding pixels in the template image. This metric is the square of the $$l^2$$ norm of the difference image. The general SSD distance metric becomes:

$$d_2(I_j,T)={\displaystyle \sum _{i=1}^n{\left|I_{i,j}-T_i\right|}^2}$$

Maximum Absolute Difference (MaxAD)

The MaxAD metric is also known as the Uniform Distance metric. It sums the maximum of absolute values of the differences between pixels in the original image and the corresponding pixels in the template image. This distance metric provides the $$l^{\infty }$$ norm of the difference image. The general MaxAD distance metric becomes:

$$d_{\infty }(I_j,T)=\underset{x\to \infty }{\lim }{\displaystyle \sum _{i=1}^n{\left|I_{i,j}-T_i\right|}^p}$$

which simplifies to:

$$d_{\infty }(I_j,T)=\underset{i}{\overset{n}{\max }}{\left|I_{i,j}-T_i\right|}^p$$

References

[1] Koga T., et. Al. Motion-compensated interframe coding for video conferencing. In National Telecommunications Conference. Nov. 1981, G5.3.1–5, New Orleans, LA.

[2] Zakai M., “General distance criteria” IEEE Transaction on Information Theory, pp. 94–95, January 1964.

[3] Yu, J., J. Amores, N. Sebe, Q. Tian, "A New Study on Distance Metrics as Similarity Measurement" IEEE International Conference on Multimedia and Expo, 2006 .

Extended Capabilities

C/C++ Code Generation
Generate C and C++ code using Simulink® Coder™.

Fixed-Point Conversion
Design and simulate fixed-point systems using Fixed-Point Designer™.