Machine learning-based edge detector

Question

I have Read the following blog on edge detection using machine learning. They

used a modern machine learning-based algorithm. The algorithm is trained on images where humans annotate the most significant edges and object boundaries. Given this labeled dataset, a machine learning model is trained to predict the probability of each pixel in an image belonging to an object boundary.

I would like to implement this technique using opencv.

Does anybody have an idea or know how this method can be implemented/developped using Opencv ?

How can we annotate the most significant edges and object boundaries for use with machine learning algorith ?

Answer 1

There are strong edge detection algorithms on opncv. The famous on is Hough transform (conjuction of lines which is described on the blog). Most of the strong edge detectors are based on gradient; gradient in x or y or both direction. I want to introduce you Sobel edge detector and Laplacian. both are provided in opencv. For the case that you want problem is tricky and depends on your date set. There are many papers published for this important problem among those I provide you the following links:

1. Dollar slides at U. Toronto

2. Review Paper from U. twente

3. Github repository s9xie

I hope this helps you.