Matching of a template and an image.
The operator exhaustive_match matches ImageTemplate and Image within the region of interest RegionOfInterest. Hereby the ImageTemplate will be moved over all points of Image which lie within the RegionOfInterest. With regard to the parameter Mode, a matching criterion will be calculated. The result values will be stored in ImageMatch.
The following matching criteria (Mode) are available:
'norm_correlation'
------------------
ImageMatch[i][j] =
SUMu,v {Image[i-u][j-v] * ImageTemplate[l-u][c-v]}
255 * -----------------------------------------------------------------------
SQRT{ SUMu,v {Image[i-u][j-v]^2} * SUMu,v {ImageTemplate[l-u][c-v]^2} }
whereby X[i][j] indicates the grayvalue in the ith column and jth
row of the image X. (l,c) is the centre of the region of
ImageTemplate.
u and v are chosen so that all points of the template will be
reached, i,j run accross the RegionOfInterest.
At the image frame only those parts of ImageTemplate
will be considered which lie inside the image (i.e. u and v will
be restricted correspondingly). Range of values: 0 - 255 (best fit).
'dfd'
-----
Calculating the average "displaced frame difference":
ImageMatch[i][j] =
SUMu,v {ABS { Image[i-u][j-v] - ImageTemplate[l-u][c-v] }}
----------------------------------------------------------
AREA ( ImageTemplate )
The terms are the same as in 'norm_correlation'. AREA ( X ) means the
area of the region X. Range of value 0 (best fit) - 255.
To calculate the normalized correlation as well as the ``displaced frame
difference'' is (with regard to the area of ImageTemplate)
very time consuming.
Therefore it is important to restrict the input region
(RegionOfInterest if possible, i.e. to apply the filter only
in a very confined ``region of interest''.
As far as quality is concerned, both modes return comparable results, whereby the mode 'dfd' is faster by a factor of about 3.5.
|
Image (input_object) |
image -> object : byte |
| Input image. | |
|
RegionOfInterest (input_object) |
region -> object |
| Area to be searched in the input image. | |
|
ImageTemplate (input_object) |
image -> object : byte |
| This area will be ``matched'' by Image within the RegionOfInterest. | |
|
ImageMatch (output_object) |
image -> object : byte |
| Result image: values of the matching criterion. | |
|
Mode (input_control) |
string -> string |
| Desired matching criterion. | |
| Default value: 'dfd' | |
| List of values: 'norm_correlation', 'dfd' | |
read_image(Image,'monkey') disp_image(Image,WindowHandle) draw_rectangle2(WindowHandle,Row,Column,Phi,Length1,Length2) gen_rectangle2(Rectangle,Row,Column,Phi,Length1,Length2) reduce_domain(Image,Rectangle,Template) exhaustive_match(Image,Image,Template,ImageMatch,'dfd') invert_image(ImageMatch,ImageInvert) local_max(Image,Maxima) union1(Maxima,AllMaxima) add_channels(AllMaxima,ImageInvert,FitMaxima) threshold(FitMaxima,BestFit,230.0,255.0) disp_region(BestFit,WindowHandle).
If the parameter values are correct, the operator exhaustive_match returns the value 2 (H_MSG_TRUE). If the input is empty (no input images are available) the behaviour can be set via set_system('no_object_result',<Result>). If necessary, an exception handling is raised.
exhaustive_match is reentrant and automatically parallelized (on tuple level).
Image filters
