VI User Manual - 第252页

Tools library 7 - 90 Vision 2007 4.10 User Manual Re v 01 The Search tool also uses 3 accura cy levels: coarse, fine, very fine. 7.15.1.4 Algorithms and performance Choosing an a lgorithm an d an accuracy leve l can grea…

Tools library

Vision 2007 4.10 User Manual Rev 01 7 - 89

7.15 Search

7.15.1 Search tool definition

The Search tool detects and locates previously trained patterns in the image. It identifies

groups of pixels of the search area that are similar to the trained model, using normalized cor-

relation.

7.15.1.1 Applications

The Search tool is mainly used to find pin tips for backplane or connector inspection.

7.15.1.2 Results

It returns presence, X, Y and Theta position of the best matching shape.

7.15.1.3 Search tool features

The Search tool can use 4 distinct algorithms to find instances of the model.

CNL normalized

When you perform a search using CNL normalized linear mode, it returns the loca-

tion of the part of the search image with pixel values that are the most closely corre-

lated to the pixel values in the model image. This type of searching is called intensity

correlation searching because the degree of similarity between the search image

and the model image is determined by calculating the correlation coefficient be-

tween the patterns of gray-scale pixel values in the two images. The method used

to compute the correlation coefficient between the two images is not affected by lin-

ear changes in brightness between the images.

CNL non linear

When you perform a search using CNL Search in nonlinear mode, it returns the lo-

cation of the part of the search region with the pattern of edges that most closely re-

sembles the pattern of edges in the model image. Because CNLSearch’s nonlinear

mode searches for patterns of edges instead of patterns of pixel values, CNLSearch

is immune to both linear and nonlinear brightness changes between the model im-

age and the search image, as long as the brightnesschanges do not affect the pat-

tern of edges in the search image.

When using non linear detection, you can specify the parameters of the edge detec-

tion (upper and lower thresholds). The edge thresholds set the edge strength (the

difference in pixel values across the edge) that CNL non linear uses to identify an

edge. All edges with strengths above the high threshold are included in the edge

map that will be compared to the runtime image. All edges with strengths below the

low threshold are excluded from the edge map. Edges with strengths between the

thresholds are included in the edge map if they are connected to another edge from

the edge map, either directly or through other edges with strengths between the

thresholds.

Normalized search

This algorithm is similar to CNL normalized but uses a more aggressive approach to

locating likely matches. Because of this, it may tend to discard some unpromising

locations prematurely.

Absolute search

This algorithm is similar to Normalized search, but the scoring function is based on

the correlation coefficient. It ignores the gray levels, considers only absolute con-

trast, and so accepts perfect mismatches as well as perfect matches.

Tools library

7 - 90 Vision 2007 4.10 User Manual Rev 01

The Search tool also uses 3 accuracy levels: coarse, fine, very fine.

7.15.1.4 Algorithms and performance

Choosing an algorithm and an accuracy level can greatly affect the tool’s processing

time. The following tables indicate the relative speed difference for using the differ-

ent search methods.

Relative search times, score greater than confusion threshold

Relative search times, score less than confusion threshold

7.15.2 Using the Search tool

There are 2 steps to use the Search tool. First you need to train a model, then you can find

instances of that model.

Training the Search tool, you have to select the algorithm(s) and the accuracy level(s) that

you will be able to use at run time.

Running the tool you can choose one of the trained algorithms and select the accuracy level.

The decision parameters are the acceptable score, the confusion threshold, and the accept-

able contrast.

The Search tool uses the confusion threshold and score that you supply to ensure that the

correct instance of the model within the search image is located as quickly as possible. The

confusion threshold is the more important to obtaining good results.

The tool uses both the score and the confusion threshold when considering whether or not

a match represents a valid instance of the model. The confusion threshold is the score above

which any match is guaranteed to be an instance of the model; all matches with scores great-

er than or equal to the confusion threshold are considered to be valid. The acceptable score

is the lower score limit of all valid matches. But other matches, which might not be actual in-

stances of the model, can receive scores above this acceptable score.

The tool uses the confusion threshold to speed the search process. If you are searching for

a single instance of the model in an image, as soon as the Search tool finds an instance with

a score above the confusion threshold, it stops searching and returns the location of the

match. If Search does not find a match with a score above the confusion threshold, it locates

all the matches with scores above the acceptable score and returns the location of the match

with the highest score.

Algorithm/Accuracy level

Coarse Fine Very fine

Cnl normalized

± 2 pixels ± 1 pixel ± 0.25 pixel

Cnl Nonlinear ± 2 pixels ± 1 pixel ± 0.5 pixel

Normalized Search ± 2 pixels ± 1 pixel ± 0.25 pixel

Absolute Search ± 2 pixels ± 1 pixel ± 0.25 pixel

Algorithm/Accuracy

Coarse Fine Very fine

Cnl normalized

100 % 110 % 120 %

Cnl Nonlinear 100 % 150 % 200 %

Normalized Search 100 % 110 % 120 %

Absolute Search 100 % 110 % 120 %

Algorithm/Accuracy

Coarse Fine Very fine

Cnl normalized

100 % 150 % 200 %

Cnl Nonlinear 100 % 200 % 300 %

Normalized Search 100 % 150 % 200 %

Absolute Search 100 % 150 % 200 %

Tools library

Vision 2007 4.10 User Manual Rev 01 7 - 91

You should set the confusion threshold high enough to ensure that confusing features in a

search image do not receive scores above the acceptable score.

The contrast threshold is the ratio of the pixel values dispersion from the trained image to the

real image. It is relative to the trained image.

7.15.3 Model description tab

1. On the Model description tab, in the Edit a model window, choose the processing:

Search. A new tab appears (with the tool name) behind the Model description tab.

2. Click Auto area to place the model encompassing area exactly on the image.

7.15.4 Search tab

Use

Realign this processing

(

) section if you want to realign

the position of the tool with another

one (not available for window 1).

2. In Position & Size search

window (mm)

(B) section, enter

the size and position of the

search window (zone in which

the feature will be searched for).

3. In Light level (C) section, se-

lect the light level you want to

use.

4. In Parameters (D) section, en-

ter the Search tool parameters

(see below

7.15.4.1 Search pa-

rameters

5. Click Test (E) button to test the

Search tool and display the re-

sults (see below

7.15.4.2 Search

test

7.15.4.1 Search parameters

Configuration tab

Click Configuration tab

to access the main run

time parameters.

1. In Fonctional param-

eters

section:

Select the

Algorithm

(A) that will be used,

among the trained algo-

rithms.

Select the

Accuracy (B)

level from the trained accuracy levels.

If you need a special

equation to realign, you

can do so using the

Equations button.