5. OpenCV Tutorial - Learn Classic Computer Vision & Face Detection (OPTIONAL)/13. Scaling, Re-sizing and Interpolations - Understand How Re-Sizing Affects Quality.srt

1
00:00:01,060 --> 00:00:08,920
So moving on to a simple fine transmission is resizing also called scaling and resizing you mating is

2
00:00:08,950 --> 00:00:09,820
actually quite simple.

3
00:00:09,820 --> 00:00:15,670
However it involves something called intercalation and interpellation is basically when we resize an

4
00:00:15,670 --> 00:00:15,970
image.

5
00:00:15,970 --> 00:00:19,690
Imagine your expanding points or pixels.

6
00:00:19,930 --> 00:00:23,960
So when you resize an image from big from small to big.

7
00:00:24,160 --> 00:00:27,740
How do we find the gaps in between the pixels.

8
00:00:28,060 --> 00:00:33,490
So that's what interpellation I have just drawn this simple little diagram here to show you what Gippi

9
00:00:33,560 --> 00:00:37,510
a square and it's because when you're doing G-B obstructing you actually interplay at points quite a

10
00:00:37,510 --> 00:00:38,330
bit.

11
00:00:38,350 --> 00:00:41,340
So imagine the red points of pixels.

12
00:00:41,590 --> 00:00:43,540
And these are the actual pixels here.

13
00:00:43,580 --> 00:00:49,750
Now imagine previously for these pixels we're closer together or red pixels ignore yellow pixels and

14
00:00:49,750 --> 00:00:50,970
every pixel for now.

15
00:00:51,370 --> 00:00:53,780
So we just haven't read pixels that are closer together.

16
00:00:53,980 --> 00:00:59,560
But no we've stretched them out and we need to approximate the pixels in-between that's when into position

17
00:00:59,560 --> 00:01:05,380
as in depletion as the methods we use to actually put generate these pixels here.

18
00:01:05,530 --> 00:01:13,160
So as kids can see the CBS step example going up in the straight line not putting a pixel here a fairly

19
00:01:13,160 --> 00:01:13,700
good idea.

20
00:01:13,710 --> 00:01:15,050
Fairly good idea.

21
00:01:15,060 --> 00:01:19,630
However doing this linear linear meaning straight line joining between pixels.

22
00:01:19,880 --> 00:01:25,860
We actually don't reflect that maybe the vehicle took a corner here instead of equal to maybe some stretches

23
00:01:25,890 --> 00:01:28,740
of buildings here in this dome example.

24
00:01:29,390 --> 00:01:35,640
And that's a form of interpellation which which we call linear interpolation.

25
00:01:35,850 --> 00:01:42,870
So let's talk a bit about the types of into positions here is interior nearest linear Kubik and Lanzas

26
00:01:43,840 --> 00:01:45,240
for apparently.

27
00:01:45,660 --> 00:01:49,910
So these are all methods and I won't go into details here it is actually a very good comparison on this

28
00:01:49,920 --> 00:01:50,790
link here.

29
00:01:51,120 --> 00:01:57,270
But generally into the area into here it is good for shrinking or sampling images.

30
00:01:57,270 --> 00:02:03,390
That's actually another form of into position as well probably just what it does is images the pixels

31
00:02:03,390 --> 00:02:04,240
was in the area.

32
00:02:04,380 --> 00:02:10,790
That's why it's good for downsampling into Nereus as fast as up or down according to this comparison

33
00:02:10,800 --> 00:02:18,150
here is good as in this case is a linear interpolation or simply good and scientific method use zooming

34
00:02:18,430 --> 00:02:23,430
all of it into Kubic which is probably what's going to generate this green point here is definitely

35
00:02:23,430 --> 00:02:26,450
better and Lanzas or luck.

36
00:02:26,500 --> 00:02:28,610
This is actually the best.

37
00:02:28,710 --> 00:02:29,900
So you can check them out here.

38
00:02:30,080 --> 00:02:32,200
So let's actually implement this in some code now.

39
00:02:34,900 --> 00:02:41,260
So let's now implement some simple scaling or resizing of NCB so you can see the uses a CV to resize

40
00:02:41,260 --> 00:02:43,540
function when the arguments listed up here.

41
00:02:43,810 --> 00:02:48,820
But as actually use this function to get a feel of it sufficiently the LoDo image and then using the

42
00:02:48,820 --> 00:02:52,010
resize function it takes all input image none.

43
00:02:52,060 --> 00:02:58,280
Which all this gustily to one shortly f x and f y These are the factors we want to scale.

44
00:02:58,390 --> 00:03:04,570
And what we're telling it is if we want to Scioto image by trick What is so true to the originals original

45
00:03:04,570 --> 00:03:05,530
size.

46
00:03:05,680 --> 00:03:12,120
And the reason we said it at some point 75 percent favor here is that we want to Munteanu image issues.

47
00:03:12,190 --> 00:03:19,810
So by default they said it uses it into linear as its assets into politian matter.

48
00:03:19,850 --> 00:03:25,470
We're actually going to try into Kubic into area and we're actually going to change the dimensions here.

49
00:03:25,490 --> 00:03:26,210
We were size 2.

50
00:03:26,240 --> 00:03:29,870
So this actually allows us to resize exactly the dimensions we want.

51
00:03:29,970 --> 00:03:32,840
So you understand that shortly when we actually implement this code.

52
00:03:32,870 --> 00:03:39,610
So let's run this Corneau as you can see this is a picture of the loop they took and it's actually smaller

53
00:03:39,700 --> 00:03:42,760
than the virtual images by truck with its eyes.

54
00:03:43,300 --> 00:03:47,710
This one is much bigger This is twice the size and this is using the entire cubic which is actually

55
00:03:47,710 --> 00:03:49,420
a slower method of interpellation.

56
00:03:49,660 --> 00:03:52,910
However the resizing looks decent quality as well.

57
00:03:53,440 --> 00:03:54,970
And this is the one that's skewed.

58
00:03:55,030 --> 00:04:00,260
So we actually set the dimensions here and let's bring this into this view here.

59
00:04:00,610 --> 00:04:03,670
So we actually said the width to be 900 and the height to be 400.

60
00:04:03,730 --> 00:04:12,550
As I previously showed in the code here we similarly we can actually skew in one dimension or see resizing

61
00:04:12,820 --> 00:04:14,190
differently in one dimension.

62
00:04:14,470 --> 00:04:19,520
So if we implement something that looks like this year which is sort of compressed horizontally here.

63
00:04:20,200 --> 00:04:20,610
OK.

64
00:04:20,650 --> 00:04:22,180
So that's it for resizing.

65
00:04:22,180 --> 00:04:24,100
It's actually a pretty basic function.

66
00:04:24,370 --> 00:04:26,050
So why not move on to some other topics.

67
00:04:26,230 --> 00:04:26,700
Inoffensively.