"We offer new support options and therefor the forums are now in read-only mode! Please check out our Support Center for more information." - Vuforia Engine Team

how to create optimal target

By Jason on Sat, 07/13/2013 - 14:40

Hi,  I am testing cloud recognition with my own target.   My targets are created by scanning the prints.  I am using 600dpi and the size is scaled to around 1024 x 800.  After creating the targets, I found the target images were very much blurred and quality usually rates under 2 stars.  Thus the cloud recognition result were very disappointing.   Sometimes, it takes around 1 minute to be recognized.  Sometimes, it simply fails recognizing.   I doubt this might relate to my method on generating targets.   So I would like to know

1) what is the optimal resolution for a target?  which one usually gives better recognition result giving same textures, 400x360, 800x600 or 1024 x800?

2) how much dpi (300, 600 or 1200) should be used when scanning prints?

3) what image processing is required after scanning before creating the target?

Thanks.

AlessandroB

Sun, 07/14/2013 - 07:06

Permalink

Hi, if you scan a printed image, you should try to make sure that relevant "features" are not removed from the image itself;

this can also be affected by the quality of the printed images, so it is hard to tell what is the right DPI to use, or the ideal resolution;

Hi, Alessandro,

I cannot change the printed image as it is already done.  The only way I can get the image is by scanning.  Should I keep the scanned image untouched or I should apply some processing to strengthen contrast before createing target?

 

Hi, Alessandro, I did a test which leaves a little bit confused over the difference of cloud recognition and on device recognition.

Hi, Alessandro, I did exactly what your described.   Since I test on my own datasets, I modified the cloud reco sample a little bit.  It no longer fatch book info any more.  The recognition process stops when onNewSearchResult is called.

There is a look-up being performed against your Cloud Database in which the camera image is submitted to the service for analysis and the search result is returned to the device. This round trip is necessarily going to take a bit longer than detecting the same target in a local Device Database.