machine vision

I need a way of recognizing a unique irregularly shaped object within an image. I've tried using some template matching techniques but this does not work in all situations. Sometimes the object is painted differently or there is a lot of noise. Also, the object rotates and moves. I need an algorithm that is based on finding the object's shape. Any help would be appreciated. Especially if the source code already exists somewhere and all I have to do is input the template image and main image

Answer 1

[Image Analyst]

Uh, yeah. There's a million image segmentation algorithms out there with varying degrees of success. You'll have to narrow down your parameters some otherwise you'll get nowhere. Let's say I want you to find "me" in the photo. I'm usually wearing different clothing. Sometimes just a brightly colored swimsuit, other times a dark trenchcoat. Varying amount of skin showing. Sometimes I have a hat or facial hair, sometimes not. Sometimes I'm facing the camera, sometimes not. Sometimes all you can see is the top of my head. In many of the images I'm in a crowd of dozens of other people. Sometimes the image is dark and noisy or blurred. Can you find me? It would help if you post some images somewhere of the best case and several of the worst case or wildly different cases. Basically we'd need to see the entire range of images you expect to encounter, along with an explanation of what the object is and where it is in each photo.

Answer 2

[John]

Let's just say the object I am trying to find is a 20 oz soda bottle in an outdoor scene. It may be a different brand of soda. The shape is never distorted but it may be rotated or moved. Also it's size may appear to change slightly because of the perspective view. Also the video may be shot in different weather conditions and either during the day or at night.

Answer 3

[John]

In my search to figure out this problem, I ran across something called the Hausdorff Distance that I thought might be applicable in this situation. It seems to compare images based on their shape. I'm not sure how to implement this in Matlab though. Could someone provide me some code snippets or pseudo code as to how this might be implemented? Thank you.

[Walter Roberson]

http://www.mathworks.com/matlabcentral/fileexchange/26738