5. OpenCV Tutorial - Learn Classic Computer Vision & Face Detection (OPTIONAL)/28. Matching Contour Shapes - Match Shapes In Images Even When Distorted.srt

1
00:00:00,430 --> 00:00:00,840
OK.

2
00:00:00,930 --> 00:00:04,890
So let's open up a four point four which is matching Contos shape.

3
00:00:05,050 --> 00:00:07,810
So that's open it here really have it open.

4
00:00:07,890 --> 00:00:14,500
So shape matching shape matching is actually a pretty cool technique implemented in open C.v.

5
00:00:14,700 --> 00:00:18,500
So the code looks a little bit confusing but it's actually quite simple.

6
00:00:18,570 --> 00:00:20,310
So let's run the code and see what's going on.

7
00:00:20,310 --> 00:00:27,000
Fiercly So we have this is a ship template image dubber trying to find in a lot of image.

8
00:00:27,320 --> 00:00:28,020
And here we go.

9
00:00:28,020 --> 00:00:30,430
So this is the image we're trying to find it here.

10
00:00:30,840 --> 00:00:35,130
And we have identified a sheep here is more similar to the shape here.

11
00:00:35,460 --> 00:00:40,770
Alice according to the algorithm there are different methods of doing the proper approximations or ship

12
00:00:40,770 --> 00:00:41,160
matching.

13
00:00:41,160 --> 00:00:42,770
I should say so.

14
00:00:42,810 --> 00:00:44,380
This one here works pretty well.

15
00:00:44,400 --> 00:00:46,150
It actually found a close match.

16
00:00:46,440 --> 00:00:48,640
So now let's see what's going on here.

17
00:00:48,990 --> 00:00:53,360
So as I said we load a template image and here that's a four star.

18
00:00:53,640 --> 00:01:00,330
And that's cool a template image and then we actually have luto a target image here which is the focal

19
00:01:00,450 --> 00:01:01,540
shapes to match.

20
00:01:01,650 --> 00:01:05,070
Or you can consider it to be much too.

21
00:01:05,110 --> 00:01:05,660
All right.

22
00:01:05,660 --> 00:01:10,760
And then what we do next is we find floozie Trishul those images here.

23
00:01:11,000 --> 00:01:15,680
That's part of the procedure that we use in this in this code here.

24
00:01:15,800 --> 00:01:19,730
And then after that we find CONTO is the first template image.

25
00:01:19,730 --> 00:01:25,820
So we extract all Kontos into this image then we sought the Contos and largest A small This just in

26
00:01:25,820 --> 00:01:31,430
case there were any noisy contours or discrepancies because we do know we only want the largest Cantal

27
00:01:33,310 --> 00:01:37,560
and necks since we have sorted Contos by in order from largest the smallest.

28
00:01:37,870 --> 00:01:41,110
We actually remember previously when there's a white background.

29
00:01:41,210 --> 00:01:45,520
The First Consul was always the big box big white box of the image frame.

30
00:01:45,610 --> 00:01:47,050
You can consider it.

31
00:01:47,170 --> 00:01:52,490
So we actually have to get a second largest contour here so we extract Descanso now and then what we

32
00:01:52,510 --> 00:01:58,700
do we actually again find Contos now in our target image which was a shapes to match to be matched to

33
00:01:58,890 --> 00:02:00,180
I should say.

34
00:02:00,280 --> 00:02:01,870
So we do that here.

35
00:02:02,590 --> 00:02:09,120
And then OK so then we actually know loop through all contours in our target image.

36
00:02:09,300 --> 00:02:13,570
And this is where we actually use C-v to match shapes function.

37
00:02:13,590 --> 00:02:14,550
So what we do here.

38
00:02:14,590 --> 00:02:19,710
This is a template Cantal dimply control was actually what we found previously.

39
00:02:20,630 --> 00:02:21,200
Right.

40
00:02:23,230 --> 00:02:27,240
And C see the contours in the control file.

41
00:02:27,310 --> 00:02:29,860
So we're going to target file has multiple contours.

42
00:02:29,850 --> 00:02:31,790
There are multiple shapes in that file.

43
00:02:31,870 --> 00:02:35,300
So that's where we're looping through each of the contours in that file here.

44
00:02:35,680 --> 00:02:41,600
And these parameters here which are different method and method parameters these are for first method

45
00:02:41,650 --> 00:02:47,260
it describes the CONTO matching type which will get into in a moment and then dispersement here which

46
00:02:47,260 --> 00:02:48,950
is just basically a default Perlman's.

47
00:02:49,300 --> 00:02:53,710
Don't interfere with that is probably going to be some works in the open Sivy is an open source project

48
00:02:53,710 --> 00:02:58,390
is probably going to be some work being done on disorder and um but for now it's not utilized totally.

49
00:02:58,420 --> 00:03:02,910
So just leave it at zero and it's actually a float zero apparently.

50
00:03:02,950 --> 00:03:08,160
So what this function returns is a much value.

51
00:03:08,550 --> 00:03:12,190
And basically Lua means a close a match to original image.

52
00:03:12,210 --> 00:03:19,500
If we were matching the exact contours like axium scale and size in order the match would be basically

53
00:03:19,500 --> 00:03:20,560
zero.

54
00:03:20,570 --> 00:03:26,270
However in this case we're looking for the closest match so we can try a different much method here

55
00:03:26,550 --> 00:03:28,130
which I'll get into in one second.

56
00:03:29,540 --> 00:03:35,150
So after we do so after we print the much value here what we do here is that just by trial and error

57
00:03:35,300 --> 00:03:40,890
I figured out that anything that's on that is value of point 1 5 is going to be the closest much or

58
00:03:40,890 --> 00:03:44,960
the much that's closest to the starship on the image.

59
00:03:44,960 --> 00:03:47,230
So that's how we determine whether it's a match or not.

60
00:03:47,390 --> 00:03:53,540
And if it is a match we make that CONTO that we examining in this loop here equal to the closest CONTO

61
00:03:53,540 --> 00:03:55,730
match either way as close as is.

62
00:03:55,750 --> 00:03:59,290
No are here.

63
00:03:59,890 --> 00:04:08,600
And then once that's done we actually draw close control control using draw control as an output.

64
00:04:08,640 --> 00:04:11,050
So that's pretty much how this could works here.

65
00:04:11,070 --> 00:04:12,850
So let's run it one more time.

66
00:04:12,980 --> 00:04:13,970
Real stuff.

67
00:04:14,250 --> 00:04:18,570
And this is a CONTO the Crucis much to CONTO of the star.

68
00:04:18,650 --> 00:04:22,170
So remember I said there are actually different matching methods here.

69
00:04:22,460 --> 00:04:27,660
Despina to parameter this much value saurian is much ships function.

70
00:04:27,800 --> 00:04:34,160
So if you go to open civies documentation here which I'll put a link in our file actually just to see

71
00:04:34,160 --> 00:04:42,300
if keeping make this a mock document you guys do they go.

72
00:04:42,410 --> 00:04:46,360
So if you look at this here are actually tree methods here is Contos much.

73
00:04:46,380 --> 00:04:49,530
I want to add a tree and this is the mathematics behind it.

74
00:04:49,530 --> 00:04:50,570
What's going on here.

75
00:04:50,910 --> 00:04:52,700
So feel free to experiment with it.

76
00:04:52,710 --> 00:04:56,470
We can try changing it to here and see what happens.

77
00:04:56,490 --> 00:04:57,710
It does work and.

78
00:04:57,750 --> 00:05:00,970
But you can see the values here are indeed different.

79
00:05:01,020 --> 00:05:06,270
In fact they're pretty much all under 1.5 here.

80
00:05:06,390 --> 00:05:12,390
So we actually pretty much got lucky because last one here was actually too much essentially and that's

81
00:05:12,390 --> 00:05:16,380
why Close's counter can't count or was equal to see.

82
00:05:16,380 --> 00:05:21,960
So you can play with this and see try different values that up and again here.

83
00:05:22,400 --> 00:05:23,140
So yeah.

84
00:05:23,150 --> 00:05:27,980
So that's how we match onto a ship using this much shape's function in open C-v.