Dear Vyacheslav Lyashenko,

I hope these articles are helpful to you. Also, you can using make binary, after extract contour with ImageJ.

Best wishes,

Ion.

Dear Vyacheslav Lyashenko,

The high pass filter coefficients in the wavelet filter will provide edges in an image.different wavelet filters will provide results but the computational complexity will increase in a wavelet which will have more number of coefficients

I suppose that the main question here is how to decide if the extracted contour is a good one. This further boils down to problem that even humans do not agree with the manual boundary segmentations. Also what the level of boundaries should be. I attached couple of course slides from Iasonas Kokkinos to illustrate this.

Kokkinos also referred to work where multiple instance learning was used to learn the ground truths from several annotations or methods (couple of references given in the slides). The ground truths can be then utilized, using some overlapping approach, to rank the extracted contours. Of course, one might take single human annotation and use that as ground truth if there is no need to compare between different segmentation methods, but only within different wavelet families. Further, while I haven't done this kind of work, I would  expect that there exist several segmentation databases for evaluation of different algorithm so that no more manual work is needed (Google might tell).

I hope that this is of some help. Also some more course material may be still available at: [ http://www.ee.oulu.fi/research/imag/courses/Kokkinos/ ]. And there might be also a reference to related articles like "Learning to Detect Natural Image Boundaries Using Local Brightness, Color, and Texture Cues" from David R Martin in PAMI 2004, pp 530-549.