Register to reply

Pattern recognition using splines

by Charles49
Tags: pattern, recognition, splines
Share this thread:
Charles49
#1
Jan10-13, 10:25 AM
P: 87
I am supposed to have an image consisting of overlapping curves. What I have is an image where the curves are missing certain segments of them. However with the human eye we can tell that certain segments are part of the same curve. (see attached figures). I want to know how I can write a program to connect curves with splines that visually we know are related.

In other words if I have a bunch of segments, how do I connect them with splines so that I get what my human eye is telling me is the correct match.

This is a difficult problem so any reference would also be very helpful.
Attached Thumbnails
whatIwant.png  
Attached Images
File Type: bmp whatIhave.bmp (91.2 KB, 7 views)
Phys.Org News Partner Science news on Phys.org
Experts defend operational earthquake forecasting, counter critiques
EU urged to convert TV frequencies to mobile broadband
Sierra Nevada freshwater runoff could drop 26 percent by 2100
Stephen Tashi
#2
Jan10-13, 11:38 AM
Sci Advisor
P: 3,300
Do you have a single image? Or a set of images drawn in a similar manner - say by the same person or process? Or are you looking for a method that works on all sorts of images?

There are no known methods that work on all sorts of images without your having to fiddle with adjusting them to the particular images you have.

Known methods that work on set of similar images would require some sort of tweaking and trial and error. For example, you might have to create a "training set" of images that gave the image and the correct fill-in. Then you'd need an algorithm to fit the method's parameters to the training set.

Fillling-in a curve from broken segments is, in the jargon of mathematics, an "ill posed problem". The statement of the problem does not define how to judge whether a given answer is correct or incorrect. Whether a method works or not depends on the opinion of the person evaluating it and people's opinions can vary depending on their artistic taste or what they want to accomplish once the curves are filled-in ( like detect what part of the image is a picture of a dog etc.).

I'm ducking specifics for now. I think it will be easy to find many papers on the web that show methods of filling-in broken curves. They will work on the set of images they werel developed for and they may or may not work on your images. Since this is the math section of the forum, the first thing to get straight is that you don't have a well posed problem. That's no reason to despair. For example, the simplest statement of the problem of reconstructing anatomy from MRI data is an ill-posed problem, but people develop useful images by adding conditions to the problem that are precise enough to define what a correct solution is.
Charles49
#3
Jan10-13, 12:00 PM
P: 87
Hi Stephen,

I have a single image. It is computer generated from a series of images. Since my background is not in computer science, could you please give me a few keywords that I can google to make my search for possible answers more effective? I am looking up training set now.

Charles

Stephen Tashi
#4
Jan10-13, 12:31 PM
Sci Advisor
P: 3,300
Pattern recognition using splines

Quote Quote by Charles49 View Post
e a few keywords that I can google to make my search for possible answers more effective?
computer vision curve completion

For the more general problem of missing or blocked pieces of an image:

computer vision occlusion

For a little known top-down down approach to image recognition by the researcher Mike Alter:

computer vision Upwrite

You could also use "image recognition" instead of "computer vision".
Charles49
#5
Jan10-13, 01:03 PM
P: 87
Thank you very much.
Charles49
#6
Jan10-13, 01:55 PM
P: 87
Here is a good paper: A Tangent Bundle Theory for Visual Curve Completion
Charles49
#7
Jan10-13, 01:56 PM
P: 87
by guy ben-yosef and Ohad ben shahar


Register to reply

Related Discussions
Automatic pattern recognition and data modeling Programming & Computer Science 5
Pattern recognition and machine learning problem 2.7 Engineering, Comp Sci, & Technology Homework 2
Accelerometer - Movement pattern recognition (iphone) Introductory Physics Homework 2
Pattern Recognition Programming & Computer Science 1
Calculus and pattern recognition Calculus 19