disparityMap = disparity(I1,I2) returns
the disparity map, disparityMap, for a pair of
stereo images, I1 and I2.

d = disparity(I1,I2,Name,Value) Additional
control for the disparity algorithm requires specification of parameters
and corresponding values. One or more Name,Value pair
arguments specifies an additional option.

Input image referenced as I1 corresponding
to camera 1, specified in 2-D grayscale. The stereo images, I1 and I2,
must be rectified such that the corresponding points are located on
the same rows. You can perform this rectification with the rectifyStereoImages function.

You can improve the speed of the function by setting the class
of I1 and I2 to uint8,
and the number of columns to be divisible by 4. Input images I1 and I2 must
be real, finite, and nonsparse. They must be the same class.

Data Types: uint8 | uint16 | int16 | single | double

Input image referenced as I2 corresponding
to camera 2, specified in 2-D grayscale. The input images must be
rectified such that the corresponding points are located on the same
rows. You can improve the speed of the function by setting the class
of I1 and I2 to uint8,
and the number of columns to be divisible by 4. Input images I1 and I2 must
be real, finite, and nonsparse. They must be the same class.

Data Types: uint8 | uint16 | int16 | single | double

Name-Value Pair Arguments

Specify optional comma-separated pairs of Name,Value arguments.
Name is the argument
name and Value is the corresponding
value. Name must appear
inside single quotes (' ').
You can specify several name and value pair
arguments in any order as Name1,Value1,...,NameN,ValueN.

Example: 'Method','BlockMatching',
specifies the 'Method' property be set to 'BlockMatching'.

Disparity estimation algorithm, specified as the comma-separated
pair consisting of 'Method' and the string 'BlockMatching' or 'SemiGlobal'.
The disparity function implements the basic Block Matching[1] and the Semi-Global Block Matching[3] algorithms.
In the 'BlockMatching' method, the function computes
disparity by comparing the sum of absolute differences (SAD) of each
block of pixels in the image. In the 'SemiGlobal' matching
method, the function additionally forces similar disparity on neighboring
blocks. This additional constraint results in a more complete disparity
estimate than in the 'BlockMatching' method.

The algorithms perform these steps:

Compute a measure of contrast of the image by using
the Sobel filter.

Compute the disparity for each pixel in I1.

Mark elements of the disparity map, disparityMap,
that were not computed reliably. The function uses –realmax('single')
to mark these elements.

Range of disparity, specified as the comma-separated pair consisting
of 'DisparityRange' and a two-element vector.
The two-element vector must be in the format [MinDisparity, MaxDisparity].
Both elements must be an integer. The difference between MaxDisparity and MinDisparity must
be divisible by 16. DisparityRange must
be real, finite, and nonsparse.

Square block size, specified as the comma-separated pair consisting
of 'BlockSize' and an odd integer in the range
[5,255]. This value sets the width for the square block size. The
function uses the square block of pixels for comparisons between I1 and I2. BlockSize must
be real, finite, and nonsparse.

Contrast threshold range, specified as the comma-separated pair
consisting of 'ContrastThreshold' and a scalar
value in the range (0,1]. The contrast threshold defines an acceptable
range of contrast values. Increasing this parameter results in fewer
pixels being marked as unreliable.ContrastThreshold must
be real, finite, and nonsparse.

Minimum value of uniqueness, specified as the comma-separated
pair consisting of 'UniquenessThreshold' and
a nonnegative integer. Increasing this parameter results in the function
marking more pixels unreliable. When the uniqueness value for a pixel
is low, the disparity computed for it is less reliable. Setting the
threshold to 0 disables uniqueness thresholding. UniquenessThreshold must
be real, finite, and nonsparse.

The function defines uniqueness as a ratio of the optimal disparity
estimation and the less optimal disparity estimation. For example:

Let K be the best estimated disparity,
and let V be the corresponding SAD (Sum of Absolute
Difference) value.

Consider V as the smallest SAD value over
the whole disparity range, and v as the smallest
SAD value over the whole disparity range, excluding K, K-1,
and K+1.

If v < V * (1+0.01*UniquenessThreshold),
then the function marks the disparity for the pixel as unreliable.

Maximum distance for left-to-right image checking between two
points, specified as the comma-separated pair consisting of 'DistanceThreshold'
and a nonnegative integer. Increasing this parameter results in fewer
pixels being marked as unreliable. Conversely, when you decrease the
value of the distance threshold, you increase the reliability of the
disparity map. You can set this parameter to an empty matrix [] to
disable it. DistanceThreshold must be real, finite,
and nonsparse.

The distance threshold specifies the maximum distance between
a point in I1 and the same point found from I2.
The function finds the distance and marks the pixel in the following
way:

Let p_{1} be a point
in image I_{1}.

Step 1: The function searches for point p_{1}'s
best match in image I_{2} (left-to-right
check) and finds point p_{2}.

Step 2: The function searches for p_{2}'s
best match in image I_{1} (right-to-left
check) and finds point p_{3}.

If the search returns a distance between p_{1} and p_{3} greater
than DistanceThreshold, the function marks the disparity
for the point p_{1} as unreliable.

Minimum texture threshold, specified as the comma-separated
pair consisting of 'TextureThreshold' and a scalar
value in the range [0, 1]. The texture threshold defines the minimum
texture value for a pixel to be reliable. The lower the texture for
a block of pixels, the less reliable the computed disparity is for
the pixels. Increasing this parameter results in more pixels being
marked as unreliable. You can set this parameter to 0 to
disable it. This parameter applies only when you set Method to 'BlockMatching'.

The texture of a pixel is defined as the sum of the saturated
contrast computed over the BlockSize-by-BlockSize window
around the pixel. The function considers the disparity computed for
the pixel unreliable and marks it, when the texture falls below the
value defined by:

Texture < X* TextureThreshold * BlockSize^{2}

X represents
the maximum value supported by the class of the input images, I1 and I2.

TextureThreshold must be real, finite, and
nonsparse.

Disparity map for a pair of stereo images, returned as an M-by-N 2-D
grayscale image. The function returns the disparity map with the same
size as the input images, I1 and I2.
Each element of the output specifies the disparity for the corresponding
pixel in the image references as I1. The returned
disparity values are rounded to $$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$16$}\right.$$th pixel.

The function computes the disparity map in three steps:

Compute a measure of contrast of the image by using
the Sobel filter.

Compute the disparity for each of the pixels by using
block matching and the sum of absolute differences (SAD).

Optionally, mark the pixels which contain unreliable
disparity values. The function sets the pixel to the value returned
by -realmax('single').

References

[1] Konolige, K., Small Vision
Systems: Hardware and Implementation, Proceedings of the
8th International Symposium in Robotic Research, pages 203-212, 1997.

[2] Bradski, G. and A. Kaehler, Learning
OpenCV : Computer Vision with the OpenCV Library, O'Reilly,
Sebastopol, CA, 2008.

[3] Hirschmuller, H., Accurate and
Efficient Stereo Processing by Semi-Global Matching and Mutual Information,
International Conference on Computer Vision and Pattern Recognition,
2005.