Canny Edge Detector - Computerphile

[@salaman_]

Mike's probably my favorite guest on here-I find these videos to be of a perfect technical level, including enough in-depth explanations as to give me a great understanding of how something works, yet not leave me confused from lack or overload of information. He does a great job at conveying details. Keep it up and thanks!

[@8ytan]

This is one of my favourite series of videos you've done. I hope there's more to come.

[@agostres]

+Ben F Yes. MOAR want we.

[@The_savvy_Lynx]

+Ben F Awesome series!

[@joker345172]

God bless Dr Mike Pound for helping me pass in my Image Processing class. I have a horrible professor that makes everything more complicated than it has to be, and this series of videos has helped me out a lot. Thanks, Mike!

[@JRCSalter]

5:19 This is REALLY clever. I love these vids. How we are able to edit images with nothing but manipulation of numbers is fascinating.

[@TehMaxLightning]

+John Salter Matrix: 1 Reality: 0

[@antler4979]

I'd love to see a video on normal, bump, and displacement maps, and how they work.

[@Vank4o]

Where were you five years ago, when I was studying that and was needing such a great help?! Great videos!

[@_fRiction_]

Edge Detection is much more crispier to me now. Bravo !!

[@mohammadfallah.rasoulnejad5379]

I actually learned more than in my university class. thank you. I cant believe there is someone there who know all kinds of things like deep learning, Image processing and password cracking and so on at the same time.

[@MD-pg1fh]

It's almost unCanny how well this works.

[@nicolasmariar]

your videos are trully the best if you want to dive in a subject. Im working with a canny procesor for some AI image generation and this was just what I needed to solve a few things

[@DeJayHank]

Great stuff as usual! Hadn't thought about the thresholding being hysteresis threshold, but it really makes sense now.

[@michelleteo2538]

Thanks to this video I realised I was using Canny wrongly... I always set the upper threshold to max 255 😆😆Thank u computerphile and Dr Mike for the great explanation!

[@abdullahmubushar3665]

It's amazing how these things are done.

[@AbdulrahmanMajash]

Oh yea! This refreshed the DIP course I took back when I was a college senior! Pretty much the best video explanation of Canny you can find! If anyone is experienced in Digital Image Processing, where do you suggest to go after Gonzalez's DIP book?

[@Sagar_smh]

To the nearest bar

[@gregsadler968]

Security class + vision class this semester = Dr Mike Pound being my savior

[@ej_entertain]

Such a genius !!! I hope he makes more videos on image processing and analysis.

[@martinparidon9056]

This was remarkably useful and fun to watch. Exactly what I was looking for. Thank you.

[@juliocesarcaye2946]

Thank you compurtephile! This "series" of videos on edge detection came about at just the right time, as my final year project has to do with image processing and I was quite frankly lazy and now am in a hurry to get things done before the deadline. Really saved my life!!

[@user23724]

It's clever how that thresholding works, nice video!

[@matthewforan6397]

It would have been nice to show the adjustments you talked about and their effects on the edge detection result

[@AnimeReference]

I get the feeling that this should be an hour long video, but it keeps getting cut short.

[@Runalotski]

+Jake Surname I think they record a long video an cut out parts of conversation into multiple videos, would have been nice to show some of the adjustments they where talking about.

[@ReviewMemeNSA]

+Jake Surname Iz cus if u get 2 much edge in 1 shot u'll die

[@yousorooo]

+Jake Surname An hour long video is more like a lecture on edge detection instead of a quick video for leisure purposes.

[@Rubafix989]

+Derek Leung There's no such thing as leisure !

[@Frostbite1003]

+Derek Leung I like your profile pic!

[@Yhigma]

Dr. Mike Pound does it again!

[@jordanlayton6540]

great video. explained much better than my prof.

[@jcofta]

The best series of videos on this channel ;)

[@ITR]

Lack of edgy jokes in comment section :/

[@ReviewMemeNSA]

+MMMIK13 Dw m8, I've arrived

[@Alkajatomota]

+MMMIK13 crawling in my crawl is that enough edge

[@TheGuysYouDespise]

+MMMIK13 You can't push people into edgy jokes.

[@onometre]

+MMMIK13 you say it like that's a bad thing

[@hesgrant]

+MMMIK13 yeah, the lack of edgy jokes is.....unCanny

[@inthefade]

So is this used for photoshop magic wand tool? And I imagine it is the first step for converting a bitmap image into vector? Very cool.

[@BoIoko]

This series is awesome

[@Walsh2571]

Holy shit that's so simple and smart

[@FedePerusset]

Thanks for the video, it really helped me out a lot for some of my projects! Keep it up

[@tristanboyle4450]

very nice.. @ 6:13 i'm sure this is analogous to a noise gate in the audio world...

[@usama57926]

Nice explanation

[@airatvaliullin8420]

Mike is Dr. Grime of computerphile

[@danielpihlquist2664]

Great video! Could you do a video on finding the 3D-disparity from stereo cameras? Maybe a bit advanced, I do not know how difficult it is...

[@bestdavid]

Your videos are great ! Excellent explanation

[@larissapimentel5457]

This series is awesome! moooooarrr

[@insmallportions]

Great explanations as always in this very interesting series. A nice follow up to edge detection would be image segmentation. This is perhaps a tougher topic to summarise, but Mike has been doing really well at explaining difficult topics intuitively. Also you could spread it over a few videos.

[@aikimark1955]

Now that you've introduced this topic, how about showing content-aware (seam carving) image editing? It was introduced at SIGGRAPH in 2010.

[@gilzeevi9263]

really liked the explaination! thanks

[@JacopoMTK]

Very interesting and easily put,thanks.

[@TheHoaxHotel]

For every pixel, there is an equal an opposite pixel.

[@ReviewMemeNSA]

+The Hoax Hotel Dats just another pixel u skrub

[@HubrisInc]

+ReviewMemeNSA 0x00 na it's an invert function u div

[@Booskop.]

+The Hoax Hotel That's what Isaac Newton said when he sat under a tree and a pixel fell out!

[@lukasdon0007]

+INSANE NL Didn't Newton get an apple on his head, and then say "for every apple there is an equal but opposite orange"? I seem to recall that is what he said.

[@Booskop.]

Wasn't it: "for every orange there is an equal but opposite purple"?

[@123-y3v1h]

Thank you for those informative videos!

[@amirhasanbasic6087]

You can use emgu CV for c#

[@DeJayHank]

Good good! Now Stereo Vision next! Block matching and disparity map and this n' that n' everything else!

[@Droobilicious]

Can you add a new video in this series for contour detection. Like you can do with opencv

[@anjamisimovic9214]

amazingly explained. thanx a lotttttt!!!!!!!!!!!!

[@thetrainoflife8327]

YOU ARE A GENIUS

[@DrRChandra]

I wonder, because you say you iterate over the image with one filter, then another, is it practical to combine the coefficients in that 3x3 square so that the image only needs to be scanned once? I'm not sure the maths would work out quite right if it was tried that way.

[@yarinh8417]

glad if somone will answer my qes about the local maxima,when iam cheking about some pixel does i need to check is neighbors in the grdient direction or the perpendicular direction to the grdient?

[@vvviiimmm]

This is awesome, keep it up

[@skittles6486]

Very very helpful

[@YKLUO]

such a nice video!

[@danfrederiksen1607]

It's a good starting point to structure the thinking but ultimately the truth is that these techniques are wholly inadequate against the variability of real world image content. The short answer is neural nets, the slightly longer answer is that you can hand craft visual algorithms you just need to step up the complexity and sophistication several times relative to this. Also you use vector math for angles and such, rarely cosine. Vector math is often incredibly elegant for seemingly very complex things while the cosine stuff is infinite series for even the simplest calculations. You also use vector for line representation, not y=ax+b because that can't do vertical.

[@michaelpound9891]

+Dan Frederiksen Absolutely. These videos are very much an introduction to image manipulation and machine vision. Perhaps we'll do a video on convolutional neural nets, but let's not run before we can walk :)

[@HamPuddle]

+Dan Frederiksen Well it really just depends on what you want to do with those pictures...

[@danfrederiksen1607]

+Michael Pound warp speed, Mr Zulu :) Perhaps it would be interesting to do a video on the mobileye image processing system that Tesla motors is using for autopilot these days. If you are familiar. It's impressively fast and capable although I suspect they will have to rethink everything exactly because fast doesn't have real world robustness. As is interestingly seen in user tests where the car almost kills them at time.

[@dtawnaal]

Thank you for this

[@jabelsjabels]

I'm gonna guess that the next video in the series is... Hough Transform?

[@Ninjaznexx]

"Hysteresis thresholding" Bless you.

[@VityokChotkiy]

thanks for the video! i would recommend to watch this in 0.75 speed

[@dbnvids4088]

Perfect

[@AIinAgriculture]

hi, sir.i am working on a real-time image based herbicide applicator. I will have to detect green colour(weed) from soil background and apply herbicide. My problem is under varying lightning condition how can fix the threshold value to detect green colour. How can I make a robust system that can work under different conditions?

[@LeslieSolorzanov]

What's with the last pat of this video? I want to see what happens!

[@arkansadave23]

sobel output = grayscale image, canny output= binary image?

[@kokonanahji9062]

Can you do one for Hough Transforms ?

[@karljay7473]

anyone know where to get the code or ML libs for this? Is it a part of TensorFlow or some other open source?

[@yantakeshinoda4747]

Can and how I reference this video in a Image processing for a college work I am doing right now?

[@bilalwaseem442]

WOW kia vedyo haiii!!

[@Nooneaskedforthis]

This would be great for sport to see if they scored

[@Gunth0r]

My internal sobel edge detection algorithm is having difficulties with your shirt.

[@カラスKarasu]

Is there some book(s) or online course where I can learn more of this stuff(image processing)

[@aliihsansenel]

If you mean "computer vision", Udacity have a free course by name "Introduction to Computer Vision", you must be patient in the beginning then I think you can gain a general knowledge.

[@ahmadjavaid327]

Hello! How can I detect the palm lines in android, Please guide me!.. Actually I want to let my android app, Scan the palm picture and then separate out different lines from this picture.. Please Help..

[@droctogonapus1223]

What uses are there for this?

[@lucasmontec]

can you guys take this up to object detection? =)

[@ShaunK94]

So I just learned all of this a few months ago... This would have been extremely helpful in gaining an overall understanding much quicker. Oh well haha.

[@miguelrochajr4625]

Man, I LOVED your video. I would like to put some Portuguese subtitles. Are you guys Ok if I reupload to my channel with Portuguese subtitles? Of course, I will give you ALL the credit and the video will NOT be monetized.

[@sososong3438]

its really smart

[@ffffffff963]

2:50 what IT cares about

[@krashd]

Outside of Newcastle it's known as the Decent Edge Detector.

[@SenorQuichotte]

Run it on separate colors and use CMYK not rgb

[@ishirraj8554]

my brain detected his face and voice as Hugh Laurie :)

[@turolretar]

computerphile becomes canny

[@haviduck]

canny take the edge off, and im not even sobel. on a serious note though, median is underrated

[@ThoolooExpress]

Feels like this video was cut way short.

[@JustinHallPlus]

isn't it pronounced So-Bell, not Soble?

[@AngelusDoyle]

canny are you okay

[@matrinoxtm]

“Quite a lot” was unintelligible? Weird.

[@Tuberex]

Even as a non native speaker i understood that

[@Malmosma]

canny edge or KennyS

[@jdgrahamo]

Okay?

[@ReviewMemeNSA]

+Gammel Prutte Ok

[@Anvilshock]

That's a canny edge, like!

[@akhil6095]

His accent seems to be faster than a Bugatti!!

[@mrBlagstock]

Man crush

[@israelRaizer]

is it just me or Dr. Pound kinda looks like Toby Maguire?

[@mfaraday4044]

He explained well, But i didnt understand anything.

[@guegorilla537]

I always wonder who he's talking to

[@LemonChieff]

I read "candy" I'm disappointed :(

[@bdnugget]

buy my mixtape

[@fr3dsky980]

Died in life

[@ReviewMemeNSA]

So edgy m8

[@asdfjkloe]

4th

[@ihrbekommtmeinenrichtigennamen]

Could you make a video about Directional Cubic Convolution Interpolation? It seems to use some sort of kernel convolution thing to scale up images. Would be very interesting.