Search Processing Mechanism

Reference image patterns are registered as models and then search is performed using the parts of input images that most resemble the models. The degree of similarity is represented with a correlation value, and inspection for defects and different parts being mixed in can be performed.

The search process is performed over several distinct stages.

Register a reference model.
Note
Model Status and Measurement Processing
- Measurement time and accuracy may be affected by the status of model in the following ways. Please select measurement objects that are in good condition (clean) for Model Registration.
  - In the case of large or complicated models, processing time is prolonged.
  - With extremely small models or models without features, search processing is unstable.
Perform rough search of overall measurement region.
Search for the model over the entire measurement region.
Perform additional searching near the model.

Search Detection Method

With search processing, there are two types of detection methods: Search by "Correlation (CR)" and search by "Shape (PT)".
The detection method can be selected by changing the "Rotation" settings for each processing item.

Setting item	Description
CR	Measurement for locations where there is high correlation (similar areas) between the color of the model registration image and the input image is performed. Since brightness is normalized in this operation, changes in brightness do not affect searching for correct positions.
PT	Measurement for locations where there is a high degree of similarity between the shape of the model registration image and the input image is performed. Since measurement is based on shape consistency (profile of model), positions can be measured accurately even if portions of the model are missing in images. When the rotation angle range is wide (such as full angle), a search that is faster than a "CR" search can be performed.Note, however, that this method may be less reliable than a "CR" search if the image has low contrast or blurred edges.

Setting item

Description

Measurement for locations where there is high correlation (similar areas) between the color of the model registration image and the input image is performed. Since brightness is normalized in this operation, changes in brightness do not affect searching for correct positions.

Measurement for locations where there is a high degree of similarity between the shape of the model registration image and the input image is performed. Since measurement is based on shape consistency (profile of model), positions can be measured accurately even if portions of the model are missing in images.

When the rotation angle range is wide (such as full angle), a search that is faster than a "CR" search can be performed.Note, however, that this method may be less reliable than a "CR" search if the image has low contrast or blurred edges.

Note

Basic selection flow

Search Speed

For processing items to perform search processing such as [Search] and [Classification], you can specify the search processing speed by through the model parameter items "Stab.", "Prec.", and/or "Skipping angle". If the value specified for "Stab." or "Prec." is small or if the value for "Skipping angle" is large, the processing speed can be increased since the amount of information for the models will be decreased during the search. In contrast, if the values specified for "Stab." or "Prec." are large or if the value for "Skipping angle" is small, the processing speed is slow because search is performed without the amount of model information being reduced. Specify appropriate values for "Stab.", "Prec.", and "Skipping angle" according to the measurement conditions.

Correlation value

With processing items that use search processing, judgement is performed through correlation values.

Correlation values are used to check the consistency (degree of similarity) between actual measurement images and reference model images. If portions of a measured image are missing or if shapes are different, the correlation value is lower.

Search Angle Range, Skipping Angle

These are values, based on the model registration image, that indicate the allowable rotation interval (skipping angle) and overall maximum rotation range for the model (angle range).Search is performed for objects that most resemble these acceptable models.

Example: When angle range is 15° and skipping angle is 5°

Illustration of search angle range and skipping angle

Candidate Point Level

This is the level used for finding models when searching.Images with a correlation value higher than the candidate point level are used to establish candidate points for search inspection.