1
00:00:00,780 --> 00:00:05,010
Let's take a step back and look at fine and none fine transforms again.

2
00:00:05,400 --> 00:00:11,800
So imagine you have an image like this here where we have this is clearly not fine.

3
00:00:11,880 --> 00:00:16,900
Transform off an original top down flat image here you can see the lines here.

4
00:00:16,950 --> 00:00:21,510
They're not parallel anymore and they're actually going to join in some feature spot maybe some way

5
00:00:21,510 --> 00:00:21,930
up here.

6
00:00:21,930 --> 00:00:29,310
FLATOW So what it means is that if we have four points here we can actually use an open civic function

7
00:00:29,790 --> 00:00:35,340
to actually watch this transfer this image here and get an image like this.

8
00:00:35,370 --> 00:00:41,100
So clearly we can take a skewed image once we're aware of the four points of the corner and get back

9
00:00:41,100 --> 00:00:43,220
to your original image.

10
00:00:43,250 --> 00:00:46,380
So let's take a look at that implementing that in some code.

11
00:00:46,630 --> 00:00:46,900
OK.

12
00:00:46,910 --> 00:00:51,020
So let's look at implementing that perspective transform that we just saw.

13
00:00:51,020 --> 00:00:55,800
So you remember we firstly need the coordinates of the four corners and the original image.

14
00:00:55,880 --> 00:00:57,550
So that's what we actually have here.

15
00:00:57,560 --> 00:00:59,390
These are the four points here.

16
00:00:59,430 --> 00:01:05,930
We're basically combining it into points 3 here comprise of these four quadrants and then so it is that

17
00:01:06,120 --> 00:01:14,420
these will be called points then we have points B which is do up which is output which is a point that

18
00:01:14,430 --> 00:01:16,330
we let's go back to the image here.

19
00:01:16,350 --> 00:01:17,980
These are the points that we want to define here.

20
00:01:17,970 --> 00:01:21,670
So that's a typical for standard size.

21
00:01:22,110 --> 00:01:25,800
We need to get the data so that it actually knows what to transform it to.

22
00:01:26,190 --> 00:01:32,460
So what we do know we actually use these two points is two sets of points here with this open CV function

23
00:01:32,520 --> 00:01:34,480
called Get perspective transform.

24
00:01:34,770 --> 00:01:40,920
What this does it actually generates a matrix and this matrix here is is basically the transform that

25
00:01:40,920 --> 00:01:45,030
basically transforms points to points B.

26
00:01:45,270 --> 00:01:51,480
And then we use what perspective that function you may have seen previously to actually generate the

27
00:01:51,480 --> 00:01:57,270
or to use this sort of matrix here and generate the output image to be called warped here.

28
00:01:57,660 --> 00:02:00,600
And this is the final size of the image we want to see.

29
00:02:00,600 --> 00:02:02,860
So as you can see it's actually lines up with this here.

30
00:02:02,910 --> 00:02:07,020
These would have four coordinates of the output image that we wanted.

31
00:02:07,020 --> 00:02:15,690
So let's run this good and voila we get the actual warped page here.

32
00:02:16,260 --> 00:02:20,700
If you wanted to do something smarter you'd probably write an algorithm that gets to CONTO of this image

33
00:02:21,150 --> 00:02:26,550
that automatically gets the locations of the corner points then feeds it in and then we run everything

34
00:02:26,580 --> 00:02:27,840
automatically.

35
00:02:27,870 --> 00:02:31,950
It's actually not too hard to build is actually a low some blogs online that actually do it for you.

36
00:02:33,250 --> 00:02:37,410
So we've just seen how we generated Imitrex for none I transform.

37
00:02:37,500 --> 00:02:41,060
But what about affine transform that should be simpler right.

38
00:02:41,070 --> 00:02:46,900
And indeed it is you only need tree coordinates to actually generate the find transform.

39
00:02:46,900 --> 00:02:49,240
So similarly we have points in B.

40
00:02:49,420 --> 00:02:49,830
We are.

41
00:02:49,860 --> 00:02:52,310
We actually use a different image here and there's a reason for that.

42
00:02:52,320 --> 00:02:53,780
Also we use shortly.

43
00:02:54,100 --> 00:02:58,240
So we have tree coordinates for points in b tree coordinates for Point B.

44
00:02:58,240 --> 00:03:05,050
We use get affine transform as opposed to get perspective transform and we similarly implement those

45
00:03:05,050 --> 00:03:11,020
transforms using Wolpoff when we just switched to see I would call them Andrews which is similar to

46
00:03:11,020 --> 00:03:13,350
using the actual dimensions here.

47
00:03:13,660 --> 00:03:18,310
We just take it directly from the image and let's run this code here.

48
00:03:18,670 --> 00:03:20,500
So as you can see these are two

49
00:03:23,530 --> 00:03:25,380
ships that we wish to transform.

50
00:03:25,400 --> 00:03:28,190
So let's press any key and continue.

51
00:03:28,190 --> 00:03:35,180
And this is the affine transform as you can see it as I find transform actually rotated the image slightly

52
00:03:35,720 --> 00:03:41,840
didn't just scale anything but as you can see clearly all the lanes Montie and parallelism here.

53
00:03:42,200 --> 00:03:45,510
So that's one of the main points of transforms.

54
00:03:45,590 --> 00:03:51,500
So I hope you get a feel of using these functions here are actually quite handy and quite important

55
00:03:51,500 --> 00:03:53,930
when doing more complicated stuff in open C-v.