image processing using template matching approach

Question

i am working on a project related to template matching image processing , i have done the matching algorithm but the problem i am facing that , template matcher always yeilds the best co-relation matched in source image of template image but i want to notify or respond only when the desired output comes neither on false output. i want to serially communicate MATLAB code with arduino board UNO R3 which will generate the pulse birectionaly when the output comes ? so what should i supposed to do ? this is the code :

cam=videoinput('winvideo',2,'YUY2_320x240');


start(cam);
preview(cam);
set(cam,'ReturnedColorSpace','RGB');
get=input('get frame ???');

frame=getsnapshot(cam);
imwrite(frame,'got.jpg');

I=imread('D:\Template matcher\got.jpg');
H_Eq=vision.HistogramEqualizer;
Temlate_matcher=vision.TemplateMatcher;
Temlate_matcher.Metric='Maximum absolute difference';
Temlate_matcher.OutputValue='Metric matrix';
marker_inserter=vision.MarkerInserter('Size',30,'Fill',false,'FillColor','White','Opacity',0.75);
I=rgb2gray(I);
I=step(H_Eq,I);
Template1=imread('D:\Template matcher\ge.jpg');


Template1=rgb2gray(Template1);


H_Eq=vision.HistogramEqualizer;
Template1=step(H_Eq,Template1);


Location1=step(Temlate_matcher,I,Template1);

 marker_inserter.Shape='Square'
 output_image=step(marker_inserter,I,Location1);

figure();imshow(output_image);

Answer 1

As we have discussed in our comments, template matching using the Computer Vision toolbox will produce the best template match in the frame. However, this does not necessarily mean that the object you are searching for is located within the location of where the template best matched.

As such, what I would recommend you do is take a look at what the metric gives you for that template. In your case, you're using the maximum absolute difference. If this metric is less than some threshold, then that could mean that the template was found in the frame you are examining. If it's greater, then there is a good possibility that it wasn't. This threshold you'll have to play around with as it totally dependent on what the template looks like and the contents of the frame you're trying to look at. If it is less than some threshold, then you can send your signal to the Arduino board. Before we do this, you must change your Template Matcher setup so it looks like this:

Temlate_matcher=vision.TemplateMatcher('OutputValue', ...
                'Best match location', 'BestMatchNeighborhoodOutputPort', true);

This will allow us to get what we eventually want. As such, you can't use Metric Matrix as the OutputValue field anymore so get rid of this line in your code. Now that this is set up, you can override the behaviour of the template matching by replacing this code:

Location1=step(Temlate_matcher,I,Template1);

With this:

[Location1, NVALS, NVALID] = step(Temlate_matcher,I,Template1);

NVALID returns true if the match is fully contained within the frame and false otherwise. For your case, NVALID should always be true as the metric you have chosen is guaranteed to do matching as long as the template is contained within the frame. NVALS is the matrix of metric values that best matched the template image. In essence, the centre of this matrix gives you the best metric value that the matcher produced, and so this is the value that you want. You can cheat and just find what the minimum value is by:

val = min(NVALS(:));

As such, after this point you can check to see if val is less than some threshold. This I don't know what it would be. I guess something to play around with would be that if the best match was off by... say 5, then there may be something interesting to take a look at in the frame. As such, set thresh = 5.

Therefore, if val is less than thresh, then go ahead and signal your Arduino board.

Good luck!