View Full Version : Blob Counter
September 24, 2003, 02:15:09
I am evaluation Montsion SDK through Montivision Workbench.
I am trying to build an system that identifies if an image has a round element inside it (with a fixed size) or not.
The system i created is :
> Video Source (webcam) > AVI Decompressor > Color To Gray > Mean > Blob Counter > Video Renderer
The Mean is to equalize the ilumination.
First question : wich unity is used for the contour filter size? pixels ?
Second question: there's a way to specify that the element is rounded?
September 24, 2003, 08:05:30
Yes, the contour size filter uses pixels as unit.
To identify only objects with a special shape you have to specify a range of contour hu invariants that is accepted. Objects have to be white on black background, you can use the MV Inverse to invert a black on white image.
You can use the MV Blob Finder to define the hu invariants for your kind of objects. E.G. load a test image containing the desired shape using the MV Image File Source filter and connect it to the MV Blob Finder, then connect it to a Video Renderer. The MV Blob Finder will display the hu invariants of the found shape.
The folliwing values are calculated from a simple circle:
hu2: 2.0 e-12
hu3: 2.3 e-8
hu4: 1.6 e-11
hu5: 10 e-21
hu6: 2.3 e-17
hu7: 1.2b e-31
You can see that hu2 - hu7 are very small for a circle. The hu invariants filter in the MV Blob Counter filter would be configured like:
hu1 0.155 0.165
hu2 -0.000001 0.000001
hu3 -0.000001 0.000001
hu4 -0.000001 0.000001
hu5 -0.000001 0.000001
hu6 -0.000001 0.000001
hu7 -0.000001 0.000001
You can vary this range to fit your needs best.
September 24, 2003, 13:21:45
i final question.
What does hu means exaclty?
September 24, 2003, 14:03:00
The hu invariants are seven values calculated from moments that describe a contour. The values are independent to the image scale, rotation, and reflection except the seventh one, whose sign is changed by reflection. In practise the single values have no special quality. For simple objects the first variants are the most significant values for an object.
Powered by vBulletin® Version 4.2.2 Copyright © 2015 vBulletin Solutions, Inc. All rights reserved.