Best parameters for image recognition...

Eggplant Functional
1

What are the best parameters for image recognition? I currently have them set at the defaults which are as follows:

Image Search Time: 1.800 seconds
Precise Image Tolerance: 1
Standard Image Tolerance: 45

The reason I ask this is for this reason. I’ve built a pretty complex OCR script that recognizes via search rectangles what image is what. It’s able to do this no matter what image is behind the search rectangle. Lets say for example these are the images. An “A”, a “B” and a “C”. Now it may display “A”, “B” and “C” one time and the next time it may display “A”, “C” and “B”. Each image gets a search rectangle drawn around it, one at a time. Once the image is found, it moves to the next one. Now again these are not “Letters” as I have provided in my example. So “readtext” will not work, these are “images” of “Letters”.

Anyways, each image has different colors, different lines, different patterns on them, sometimes the pattern is the same. I have noticed recently that sometimes “A” is being recognized as a “C” or vice versa, as the pattern may be similar (but not exactly) but the color is completely different. It’s driving me bonkers at trying to figure this out. I also have my Capture Mode via the Viewer options set as follows:

Display Capture rectangle border “checked”
Detect pulsing images when capturing “checked”
Default Search Type: Text (should this be precise or tolerant maybe?)

My Platform is set to Generic OCR and my Text Engine is OCR Search.

I also have my “Connection” to my VNC (adderlink) set at “256” colors to speed up “actions”. Would increasing the color to fix this?

My Connection Type is also “Legacy (3.3)”. Would changing this to “Standard (3.8)” help?

I’m at my wits end and need help guys. This complex script works great, most the time, but there are those times when it finds things and says they are one thing, when they are not.

Oh one last thing, I have unchecked “ALL” the VNC Encodings and the only one that has been checked for a while now is “Raw”, which is grayed out and can not be unchecked.

2

Love how this just gets ignored…glad to see what our license subscriptions are really about… :evil: :roll:

3

Please submit cases requiring immediate attention to the support email address. The forums are not and have never been a primary support mechanism; they are better suited to less urgent topics that might be addressed by other users. This is in part because anyone can post to the forums and we have no way of tracking who they are or what account they belong to, since most people use pseudonyms and do not register with their work email addresses. We don’t ignore the forums, but they take a back seat to cases submitted by readily identifiable users via email, and as a result, it can sometimes take a day or more before we respond to these posts.

Submitting email to the support address automatically creates a case in our tracking system. It also makes it easier to submit screenshots and other materials and for us to reply in kind. This would be especially helpful in this case because I’m not able to picture the scenario that you are describing.

4

Its possible that increasing the color depth could help in this scenario, but it would be really helpful to see the images involved. It seems like the captured images might be small, since larger images are unlikely to generate false matches. Changing the VNC encodings can change the speed of the connection by a tiny amount, and can sometimes be used to address connection issues, but it’s not going to affect the display of the SUT or by extension the matching of images.

There are no “best” parameters for image matching – it’s entirely dependent on the images. That’s why we have options – if there was one best setting, then that’s probably all we would have.

If you can send us some sample images and screenshots we will be in a better position to advise you on the reasons behind the problems that you are seeing.