How to match two curves to find if one curve belongs to the other?

I have a video recording of moving objects that seem to be changing in size due to the fixed capturing lens position. The objects may also have slightly different shapes due to edge detection on different settings. Then the objects move and seem to change size and shape. I need to know how much each object moved (direction and translation)!

Any idea on how to compare two curves (of different lengths) to see if they share some points (with an error tolerance)?

Thanks a lot

Answer 1

[Mona]

I actually managed to find a solution to this question. Dealing with the curves is challenging. But since the curves represent objects' perimeters, it's possible to match two images instead of two curves. I found a topic on MathWorks that guided me through matching the two objects. I am sharing the link to the topic.

Find Image Rotation and Scale Using Automated Feature Matching

Many Thanks!

[Bhavesh Parmar]

Hello Mona, I am presently working on the problem, where I am tracking an object, I prepare a profile of the object movement.

For classification of the object movement I need to match the profile of movement with the predefined movement profiles in neural network (trained).

I wanted to know that, how you have matched the two curves, and which algorithm you have used?

Answer 2

[Ivan Shorokhov]

Hello Mona,

You can use following algorithm.

1. Convert two images to binary ( im2bw )

2. Use ' regionprops ', to measure area of the images.

3. Subtract two binary image form each other and measure properties one more time, that would give you difference between two images.

I hope it will help you and please vote for the answer.

Best of luck!

[Mona]

Thank you for your answer Ivan.

I don't want to find the difference in area between the two images. I want to find how much one curved moved and which part it resembles of the other curve. And eventually plot the two matched curve superimposed on each others.

Thanks a lot for your input though.