A texture is given a name. To each texture multiple reference images is associated to improve matching performance.

The tool is suitable for the following applications:

• Surface Classification
• Object Identification

Typical task for texture matching is matching images like:

Texture Matching is used and very powerful when it is impossible to extract physical features from the image.

Basic principle for Texture Matching on 8bit grey-scale images 

The image is converted to three image by applying a low - medium - highpass filter along every line in x-direction.The same filter is applied to the three resulting images in the y-direction. The nine resulting images are used to produce the nine coefficients:

  LxLy LxMy LxHy

  MxLy MxMy MxHy

  HxLy HxMy HxHy

The LxLy represents to low frequency content in x and y direction. HxLy contains the high frequency content in x and the low frequency content in y and so on.

For each image there is generated a mean value and a standard deviation for each coefficient. Experience has shown that removing LxLy which includes the DC component of the image can be advantages.

The eight dimensional vector of mean values describes a two dimensional frequency content of the image / reference. The standard deviation is describes the uncertainty in the coeefisient estimation and is used a  weight in the distance measure comparing and image and its reference.

The distance is in general described:

  sum of all abs ( RefNxNx - NxNx ) / ( RefStdev )

One can see that the coefficients with low standard deviation is weighted up and that the weight are from each reference. The standard classification is to select the reference with the smallest distance.

The distance calculation is not time consuming. This means that limited the number of references does not save anything. Adding reference will normally lead to better sorting because the class description is better the more references.

References

Reference - Reference system selection

ROI - the tool's region of interest

• Center-x - center x of the ROI
• Center-y - center y of the ROI
• dX, dY - Width and height of the search area

Textures 

• Show best reference - You can choose to show the best reference to a texture image. It is updated while running the inspection in the Selected reference image pane.
• - Adds texture with a given name
• - Removes a texture including all reference images
• - Adds a reference to the selected texture - see also "Adding references using Point & click"
• - Removes the selected reference
• Current Image - Image is updated when an inspection is performed - the image region is transferred from the captured image
• Selected Reference - displays the image of the best or selected reference
• Status bar
  • Best - name of color
  • Dist - Distance of best match
  • Sep - logarithmic separation between best texture and second best texture

Classification

Image decimation

• Active - activates image decimation

• Horizontal - horizontal decimation factor

• Vertical - vertical decimation factor

Note - Image decimation speeds up the TextureMatcher - it is by default off

Normalizing - used to make the texture matcher less sensitive to intensity variations

• Method

  • 1-Offset - default

  • 2-Offset and mean-gain

  • 3-Offset and std-gain

Features distance - specifies how to calculate the distance - difference is the best method

• Method

  • 1-Difference

  • 2-Symmetrical difference - default

  • 3-Similarity without weights

  • 4-Similarity

• Active coefficients

  • By default all coeffisient except LL is included - to deviate from default one verify the performance on a large image set

Constraints - to specify an absolute criteria to accept a texture

• Maximum distance - the maximum distance from the best reference image

• Minimum separation - the minimum logarithmic separation between the best and the second best texture

Features

Textures sorted by feature distance - The textures are here sorted after feature distance, thus you can see the best match. Reference not meeting the constraints are display with a yellow checkmark.

Visualisation

Results

ExecuteCmd support - more information

Example 1 : TextureMatcher Classification

# read the results from the texturemacther
type = GetValue('TextureMatcher.Texture')
print type
match = GetValue('TextureMatcher.Match')
print match

#classifies a set of LogicTool based on the result
SetValue('Smooth.Value',type=='Smooth')
SetValue('Tree.Value',type=='Tree')
SetValue('Structure.Value',type=='Structure')
SetValue('Pattern.Value',type=='Pattern')
SetValue('Backside.Value',type=='Backside')