5. OpenCV Tutorial - Learn Classic Computer Vision & Face Detection (OPTIONAL)/21. Dilation, Erosion, OpeningClosing - Importance of ThickeningThinning Lines.srt

1
00:00:01,660 --> 00:00:03,980
So let's explore dilution and erosion.

2
00:00:04,030 --> 00:00:09,670
These are two very important open CVF operations and are quite useful in image processing.

3
00:00:09,760 --> 00:00:12,880
So the best way to explain this is by looking at examples here.

4
00:00:12,940 --> 00:00:14,590
So this is a letter s here.

5
00:00:14,620 --> 00:00:19,660
It's a white letter on a black box all which is what the zeros and ones correspond to.

6
00:00:19,660 --> 00:00:24,600
You can consider the ones to be 255 if you want just a Munteanu open see Visa on it.

7
00:00:24,820 --> 00:00:30,440
So let's look at what erosion does erosion removes pixels at the boundaries of an object of an image.

8
00:00:30,700 --> 00:00:35,920
What it means is that the boundaries of the object the object being a letter s here.

9
00:00:36,170 --> 00:00:40,640
Imagine everything is gone here and the bargees itself.

10
00:00:40,720 --> 00:00:42,590
That's exactly what erosion does.

11
00:00:42,610 --> 00:00:49,570
So we get it out that s dilation whoever does the opposite dilution that adds pixel to the boundaries

12
00:00:49,570 --> 00:00:50,990
of an object here.

13
00:00:51,010 --> 00:00:55,210
So this is why it becomes much Tica in our dilated segment here.

14
00:00:56,200 --> 00:01:01,200
So what about opening Andalus in here opening and closing are really useful functions that combined

15
00:01:01,270 --> 00:01:03,060
dilution and intuition together.

16
00:01:03,340 --> 00:01:10,850
So opening is erosion followed by dilation and closing is the opposite dilution followed by erosion.

17
00:01:11,230 --> 00:01:16,080
As you can imagine an erosion followed by dilution would be very useful in getting rid of noise.

18
00:01:16,300 --> 00:01:19,020
So imagine if it was some little white specks in this image.

19
00:01:19,180 --> 00:01:25,720
If we Rojos images and does become Tendo to one visit or disappear altogether then we'd do a dilution.

20
00:01:25,750 --> 00:01:28,010
We actually preserve the initial image here.

21
00:01:28,020 --> 00:01:30,220
We want to maintain.

22
00:01:30,450 --> 00:01:36,150
So no dilution and erosion while simple actually causes a lot of confusion with first timers who try

23
00:01:36,150 --> 00:01:41,930
to use it at all because there's a false misconception about what it does now.

24
00:01:41,940 --> 00:01:42,540
Yes it does.

25
00:01:42,540 --> 00:01:47,640
Dilution does in fact add pixels to the boundaries of an object and erosion does remove pixels at the

26
00:01:47,640 --> 00:01:48,830
boundaries of an object.

27
00:01:49,050 --> 00:01:56,310
However imagine we have a black X on a white bacterial open even to produce this white as being the

28
00:01:56,310 --> 00:01:57,370
object itself.

29
00:01:57,600 --> 00:02:03,180
So when we run an erosion here it's actually going to erode into this you see erosion is actually going

30
00:02:03,180 --> 00:02:08,450
to have the thickening effect of dilution and the dilution is going to have the opposite effect.

31
00:02:08,460 --> 00:02:13,440
It's going to have the thinning effect that we expect and erosion to have because what's happening here

32
00:02:13,440 --> 00:02:16,900
is that the boundaries of the image are basically in the points here.

33
00:02:17,190 --> 00:02:23,580
So when we direly it we're actually making white bigger on the edges of all P and whatever else happens

34
00:02:23,660 --> 00:02:25,600
if you would here.

35
00:02:25,740 --> 00:02:27,300
So please don't get them confused.

36
00:02:27,300 --> 00:02:30,610
It happens quite often it happened to me actually.

37
00:02:30,630 --> 00:02:32,310
So don't feel bad if it happens to you.

38
00:02:33,980 --> 00:02:39,680
So let's not look at implementing dilution erosion opening and closing you know a code.

39
00:02:39,720 --> 00:02:44,380
So the first thing to note is that we actually need to define a kernel and we define it by using non-place

40
00:02:44,470 --> 00:02:46,690
ones function data type here.

41
00:02:47,070 --> 00:02:49,280
And we have a five by five.

42
00:02:49,350 --> 00:02:50,980
You know exampled.

43
00:02:52,130 --> 00:02:56,080
So to erode function and daily function both follow the same pattern here.

44
00:02:56,340 --> 00:03:01,950
We take that input image couldn't say as we defined here and iterations which is how many times you

45
00:03:01,950 --> 00:03:02,810
run it.

46
00:03:03,030 --> 00:03:05,670
In most cases you would never need to run this more than once.

47
00:03:05,670 --> 00:03:12,840
However if you do run it to a tree or whatever amount of times you actually increase effect it's running

48
00:03:12,840 --> 00:03:15,450
the erosion twice and same image.

49
00:03:15,510 --> 00:03:16,470
So let's run this.

50
00:03:16,480 --> 00:03:19,080
I mean take a look and see what happens.

51
00:03:19,080 --> 00:03:24,760
So this is the open C-v text which I wrote out in Windows beant this is it.

52
00:03:24,770 --> 00:03:25,580
It really did.

53
00:03:25,740 --> 00:03:28,810
As you can see it has the effect of what we anticipated.

54
00:03:28,830 --> 00:03:32,560
It actually erodes the boundaries here.

55
00:03:34,700 --> 00:03:36,150
Making it much tighter.

56
00:03:36,590 --> 00:03:38,000
And let's see dilution now.

57
00:03:40,880 --> 00:03:42,160
Exactly as we anticipated.

58
00:03:42,170 --> 00:03:45,770
So as you can see dilution adds pixels the edges here.

59
00:03:45,770 --> 00:03:50,910
So it actually everything appears much thicker as if it was written with a marker and not a pen.

60
00:03:51,110 --> 00:03:59,910
And that's look at opening is the effects of erosion then dilution.

61
00:03:59,930 --> 00:04:06,180
So as you can see here because just when he wrote it actually disappeared these boundaries here when

62
00:04:06,180 --> 00:04:06,770
we dilated.

63
00:04:06,770 --> 00:04:12,180
No it actually misses these things totally because there's nothing to dilate the underside here.

64
00:04:12,500 --> 00:04:16,470
So that's why opening so that's whole opening has this effect on image here.

65
00:04:16,470 --> 00:04:20,780
So let's look at closing now no closing actually looks quite nice.

66
00:04:20,780 --> 00:04:25,870
And the reason for that is because closing is dilution first then eroding.

67
00:04:25,880 --> 00:04:31,540
So what we did here actually should get us back something to a very original image and actually it does.

68
00:04:31,550 --> 00:04:38,140
If you look at it is actually pretty much to see him just barely noticeable.

69
00:04:38,180 --> 00:04:42,980
So these operations that we've just seen are actually called morphology operations and they're actually

70
00:04:42,980 --> 00:04:48,210
a few more you can take a look at this link to view the eye is open see these official documentation

71
00:04:48,250 --> 00:04:48,950
site.

72
00:04:49,130 --> 00:04:53,050
However they aren't as useful as erudition and erosion in my opinion.

73
00:04:53,090 --> 00:04:57,490
But however you may have a special uses for it so feel free to check them out.