OpenCV Python Canny Edge Detection

all right in this video we're going to talk about candy Edge detection and opencv using python so we'll start by saying what it is why do we need it how does it work and jump straight into a coding example so at the end of this video we'll see how we get this image here on the left to this image here on the right and we'll actually be making a little uh track bar application which will have some fun with the threshold values okay so what is canning Edge detection it's an algorithm that's typically more robust than standard gradient methods so a lot of times when we just do normal gradients the result might not be so good so in those cases you may have to rely on the candy Edge detection okay so why do we need it like we said is it's going to be more robust or more reliable but in general applications you might use it for things like object detection image segmentation or feature extraction okay so how does it work the way it works is typically this could be a smoothing application with gaussian filter and then the gradient of the image this can be applied using a Sobel which we talk about in image gradients and then we're going to be applying a non-maximum sub non-maximum suppression technique to find the local maximum so what this does is if you have like a thick Edge it'll transform it into a thin Edge and the idea is you know if there's if you have like a lot of Peaks there's going to be like one one Peak that's the maximum and then you just ignore the other Peaks so that's the idea with this because you could have a bunch of Peaks here and maybe only one of them corresponds to the peak so that is what non-maximum suppression is and then you have double thresholding to keep the edges that go above the max threshold even if it's connected even if connected Edge dips into region between the Max and Min threshold so so we have a couple of cases that we can look at here so um you have two thresholds typically a high and a low so if you have an edge and is all above the high threshold then we keep it okay but here we have a b and c so if you have part of an edge that's above but it's connected to another Edge that is between high and low we still keep it but if you just have an edge like e which is just between high and low and it's not passing the high threshold then we would discard e okay and then if you have something that's below the low threshold we also discard it okay so the only ones we're keeping would be these two cases okay so this we keep keep keep okay so that's the idea of the Kenny Edge detection okay so let's jump right into a coding example okay so as usual let's go ahead and read in some of our modules that we'll need so we're going to go ahead and import CV2 SCV and poor matplotlib dot Pi plots as PLT import numpy as MP and then import OS and we're going to call our function here Kenny Edge and then we have our if name equals Main and we'll call our Kenny edge here okay so inside our Kenny Edge Let's go ahead and read in our image so root equals os.get CWD and then we have our image path equals OS dot path dot join here and we'll pass in Roots we have our demo images tesla.jpg and then we have our MS equals CV dot I'm read and then we'll pass in our image path and then we want to convert BGR to RGB so we have image equals CV dot CVT color and we pass in our image CV dot color BG R to r g b okay so now we have our color image let's go ahead and rescale our image because it's a little big so we have Heights with and then the third argument which we don't care about image.shape and then what we want to do is have a scale factor we're going to get one fifth of the image and we're just scaling it because if it's too big it'll take up um it'll go out of the screen so that's the main reason so we're going to call this height scale and then we'll cast it with an INT and we have Heights times our scale then we'll repeat the same thing for the width here okay so we have our Highland with scaled and then our new image we're just going to use a CV dot resize function and we'll pass in our image and a tuple which will be our width scale and then our height scale and then our interpolation method we'll just go with CV dots enter go with the linear is fine for this okay so we have successfully resize it and we'll make do some fun with a track bar so we have a window name which we will name Kenny and then here we're going to do cv.named window and then pass in our window name and we'll create our two track bars so CV dot creates track bar and then we have the name of our track bar which will be Min threshold and we'll pass in our window name and then our value 0 and then 255. and we'll need a callback so right now our callback we're actually not going to do anything so we'll have a placeholder and we'll just call pass okay so this will be our placeholder function so we'll repeat the same thing for our Max threshold here so that will be pretty much the same and then we'll put it inside a loop so while uh true can't type right now so while true uh what we want to do is have an escape so if CV dot weights key and if that equals the q key so Q for quit we're gonna break out of this otherwise we will extract our Min threshold value using CV Dot get track bar position and the name is going to be Min Thresh and we'll pass in our window name then we'll repeat the same thing for our Max so here we're gonna just change this to Max so we have our two track bars we're getting positions of and then we'll update our CV dot um I'm show here with our candy function so we have our window name and we could go ahead and call our candy function here we'll just call this candy Edge eaglecv Dot canny so what takes it in is the image and then our Min threshold and then our Max uh threshold okay so here lastly we'll pass in our Kenny image and then all we gotta do now is just have a close with a destroy all windows here that way when we're done it'll close out so if I go ahead and run this program we should see our results so notice here we have our Min threshold and our Max threshold so if I slowly increase the max threshold you can see more of the good edges show up and some of the nonsense I just start disappearing so if I drag it all the way to the end you can see that's how it looks like now if I start playing with the Min threshold notice what happens some of the edges gets better and we could see a lot more of the car okay but notice when the car structure starts getting clearer um I'll lose some of the details of the seats right so that's kind of the drawback and you just kind of have to find a balance between what you want so here you can see a lot of the car seat detail but here it's better okay so for me I kind of like somewhere in between here maybe this is a little bit lower so something like this might be pretty good and then maybe you could apply like a secondary filter to have the results look better okay so if you found this video helpful keep a like And subscribe and I'll see you in the next one

Channel: Kevin Wood

Views: 395

Keywords: hysteresis thresholding, hysterisis, sobel, edge detector, edge detecting, canny edge detector, Edge Detection, canny edge detection, edge detection in image processing, canny edge detection opencv

Id: PS7zHKwXWRM

Length: 9min 0sec (540 seconds)

Published: Thu Jun 29 2023