Crack Detection – Segmentation

Original Image
(Original Image) – (Blurred Image) = Edge Image
Binarized Image
Color Binarized Image
(Colored Image) + (Original Image) = Highlighted Crack Image

Now this is a method that does not require any fancy machine learning or complex algorithms in python. In fact, all of this was done with off-the-shelf functions in openCV.

What if we were to apply a sliding window to detect crack regions and this basic segmentation? It would look “blocky”. The bigger the blocks the quicker the process, but the blockier the regions. Smaller the blocks, the longer the computation time, but smoother the regions. When the images coming in are 4000×6000 or are in these massive magnitudes of pixels it becomes extremely important to make these distinctions of window size, etc.

When images are taken with increasing pixel density, it becomes computationally efficient to analyze only the regions which are important. Which is why we circle back to the importance of an AI which can determine areas in an image which are more important to view than others. Thus, focusing the search on a subset of the image, rather than the entire image.

Leave a Reply

Your email address will not be published. Required fields are marked *