Pattern Recognition
[Geometry And Matching]

This group provides a package of pattern recognition functions based on standard correlation. More...

Classes
class	SoBinaryCorrelationProcessing2d
	SoBinaryCorrelationProcessing2d engine More...
class	SoGrayscaleCorrelationProcessing2d
	SoGrayscaleCorrelationProcessing2d engine More...

Detailed Description

This group provides a package of pattern recognition functions based on standard correlation.

Some tools to ease the exploitation of results are also included.

Correlation

The correlation filters allow the matching of rectangular or irregular patterns. Non-rectangular patterns are implemented with mask AOIs.

Use SoGrayscaleCorrelationProcessing2d for grayscale image correlation and SoBinaryCorrelationProcessing2d for binary image correlation.

SoGrayscaleCorrelationProcessing2d allows for local luminosity and / or contrast normalization. There are 4 different correlation types (see SoGrayscaleCorrelationProcessing2d::CorrelationMode) :

DIRECT: Direct correlation (no normalization).
MEAN: Mean normalized correlation (luminosity).
VARIANCE: Variance normalized correlation (contrast).
MEAN_VARIANCE: Mean and variance normalized correlation (luminosity and contrast).

We perform the correlation between a $N\times M$ input image, $I$ , and a $kx \times ky$ kernel, $K$ . The output image is an $N\times M$ floating point image, $O$ . The correlation coefficient at location $(i, j)$ is given by a local calculation between the model and a pattern extraction from the input image, $PATTERN$ . The pattern location is $(i-kx/2, j-ky/2)$ and its dimension is $(kx, ky)$ . The actual calculation depends on image type:

logical for SoBinaryCorrelationProcessing2d,
multiplication for SoGrayscaleCorrelationProcessing2d with MULTIPLY ,
difference SoGrayscaleCorrelationProcessing2d with SUBSTRACT or
sign change criteria SoGrayscaleCorrelationProcessing2d with SIGNCHANGE .

It also depends on the correlation normalization TYPE as shown below. When a part of the pattern lies beyond the edge of the image the correlation is not performed on the image border. SCorrelation filters provide a step parameter (see SoGrayscaleCorrelationProcessing2d::OffsetMode and SoBinaryCorrelationProcessing2d::OffsetMode) which speeds up the operation by calculating 1 value out of each step, as shown in Figure 1.

Figure 1: Standard Correlation

Luminosity and Contrast normalization

DIRECT: Direct correlation (no normalization).
$MATCHING(Pattern, K)$
MEAN: Mean normalized correlation (luminosity).
$MATCHING(Pattern-mean(Pattern), K-mean(K))$
VARIANCE: Variance normalized correlation (contrast).
$MATCHING(\frac{Pattern}{\sqrt{var(Pattern)}}, \frac{K}{\sqrt{var(K)}})$
MEAN_VARIANCE: Mean and variance normalized correlation (luminosity and contrast).
$MATCHING(\frac{Pattern-mean(Pattern)}{\sqrt{var(Pattern)}}, \frac{K-mean(K)}{\sqrt{var(K)}})$

Output Image Normalization

$dist = \frac{maxval-minval}{2}$

$med = \frac{maxval+minval}{2}$

$\begin{matrix} norm(i, j) &=& \frac{out(i, j)-med}{dist}& \mbox{by default} \\ norm(i, j) &=& -1 \times \frac{out(i, j)-med}{dist}& \mbox{for difference} \end{matrix}$

Position of the Correlation Coefficients

Figure 2: Correlation coefficient for odd dimensions

If the dimensions are even, the position of the correlation coefficient is the closest pixel position to the top and the left.

Figure 3: Correlation coefficient for even dimensions

Correlation with a Non-Rectangular Pattern

$Kern\times Mask$

$Pattern\times Mask$

$Kern\times Mask$

$Pattern\times Mask$

Figure 4: Correlation with a non-rectangular pattern

Output Data

SbCorrelationDetail

the position of the best matching (matchingPositionX, matchingPositionY),
the minimum and the maximum values of the correlation image before the conversion (minComputed, maxComputed),
the theoretical minimum and maximum correlation values before the conversion (minTheoretical, maxTheoretical).

Pattern Recognition [Geometry And Matching]

Classes