Edge detection

edit

You can try in this way:

1. Apply threholding because you need a binary image.
2. Apply morphological operator: first Open, then Close. Try with different SE.
3. Use findContours.
4. Use drawContours() with the original image.

You can try to convert to grayscale before the first threholding.

You can try to use Canny operator instead of threholding and morphogical operator before findContours().

Basically you need is a binary image as input for findContours.

Suggestion: you can also extract boundary in this way: A - (A EROSION B), where A is your image and B is your SE.

Try smoothing the image before applying the canny operator. The main idea is to keep the abrupt changes (high frequency) between the different objects, and at the same time smoothing the changes within the same object (low frequency). I think you can use a median filter or a bilateral filter (slower) for this.

Referencia: link text

I hope it helps you