Options
Asking about the criteria for determining rounded corners and sharp corners.
Hello everyone,
I have a set of samples, and after processing, the samples themselves will transition from sharp corners to rounded corners. I need to determine the actual numerical change in the object's sharp corners and rounded corners.
The problem is that existing types in software make it difficult to discern differences (e.g. roundness, diameter, etc.).
The following image is the numerical reference I used.
There are only slight differences, and there's no way to distinguish them.
As shown in the example below, the roundness of the two objects is nearly identical, but it can be observed that the object on the right belongs to a sharp-cornered type.
The main issue is how I can use actual numerical values to convince others that it has become smoother after processing.
I have a set of samples, and after processing, the samples themselves will transition from sharp corners to rounded corners. I need to determine the actual numerical change in the object's sharp corners and rounded corners.
The problem is that existing types in software make it difficult to discern differences (e.g. roundness, diameter, etc.).
The following image is the numerical reference I used.
There are only slight differences, and there's no way to distinguish them.
As shown in the example below, the roundness of the two objects is nearly identical, but it can be observed that the object on the right belongs to a sharp-cornered type.
The main issue is how I can use actual numerical values to convince others that it has become smoother after processing.
The attachment contains the measured object and its type.
Version: Image Pro 10 10.0.15
Waiting for a response, thank you, everyone.
Waiting for a response, thank you, everyone.
0
Best Answer
-
OptionsHi Kyle,
Image-Pro has a number of measurements that quantify shape, you can see them in the "Shape" category. But it might be difficult to pick one if the shapes are diverse and have some minor differences.
You can use Learning classification to determine which measurement gives you the best separation between the objects. In my test it's "Perimeter Ratio" (max weight of 0.5547). But this parameter by itself doesn't describe the class fully, it has to be used in combination with other shape parameters. The next measurement is "Fractal Dimension", which is the most suitable for your application, but the shapes must be normalized to the same area to use this parameter by itself.
Another approach is to use some steps, which are based on the nature of alternation process. For example, the corners are removed by the Open filter, so the filter will affect more the objects with sharp corners and have no effect on objects with smooth corners.
Here is the comparison of shapes after Open filter 11x11 3 passes:
Note, that the small objects are affected more by the filter. So the areas of shapes should be normalized too.
Another approach to apply smoothing is to to use Smoothing parameter of segmentation. The smoothing will affect sharp-corner objects in higher degree than smooth objects:
Then you can check the ratio of Area or Perimeter of the objects with and without smoothing to determine the class.
Yuri
0
Answers