1 00:00:00,210 --> 00:00:02,310 So now let's begin section six. 2 00:00:02,520 --> 00:00:09,300 So this one is titled Scaling Resizing into positions and cropping, so let's open this up and wait 3 00:00:09,300 --> 00:00:12,540 for it to load and should be here by now. 4 00:00:12,780 --> 00:00:13,320 There it is. 5 00:00:13,950 --> 00:00:19,530 So what we're going to do in this lesson, we're going to actually look at resizing and scaling images. 6 00:00:19,560 --> 00:00:24,210 Now that sounds like two different things, and technically it's not. 7 00:00:25,350 --> 00:00:28,230 But basically, it's just changing the size of the image. 8 00:00:28,800 --> 00:00:31,450 So it opens. 9 00:00:31,450 --> 00:00:36,480 TV has a function called CV to door to resize, and that's what we're going to take a look at first. 10 00:00:36,990 --> 00:00:42,510 So this function is quite useful, and there's many different ways to resize images which might not 11 00:00:42,510 --> 00:00:43,960 might be surprising to you. 12 00:00:44,490 --> 00:00:46,800 But we'll take a look at a few of those ways right now. 13 00:00:47,850 --> 00:00:51,420 So firstly, we have let's look at this function here. 14 00:00:51,420 --> 00:00:53,970 Let's look at what the resize to function does. 15 00:00:54,450 --> 00:00:57,390 So we have this here and highlighting that view. 16 00:00:58,020 --> 00:00:59,870 So we have the input to the image here. 17 00:00:59,880 --> 00:01:03,030 We have this size, this size, which is the output image size. 18 00:01:03,030 --> 00:01:06,960 If we want this design, some sense for dimension here. 19 00:01:07,020 --> 00:01:13,050 So we can specify actual dimensions of here, or we can have the X y scale here, as well as the interpolation 20 00:01:13,050 --> 00:01:13,440 method. 21 00:01:13,860 --> 00:01:15,330 So that could something a bit confusing. 22 00:01:15,330 --> 00:01:16,590 So let's take a look at this. 23 00:01:17,280 --> 00:01:24,180 So the interpolation methods here they are into area, into nearest, into linear, into cubic and Lonzo 24 00:01:24,660 --> 00:01:29,760 Lonzo Sorry four, which is supposedly the best resizing algorithm. 25 00:01:30,360 --> 00:01:33,000 So let's just test these out shortly. 26 00:01:33,000 --> 00:01:36,690 OK, so that actually this can happen to you guys. 27 00:01:36,690 --> 00:01:37,260 So it's a good thing. 28 00:01:37,260 --> 00:01:43,230 This lesson as I'm shooting this lesson club only allows you to open five notebooks at the same time. 29 00:01:43,740 --> 00:01:51,480 So because we're no longer using one to three and four, this just closes off too many of them and go 30 00:01:51,480 --> 00:01:52,290 back to this year. 31 00:01:53,190 --> 00:02:00,320 So now, because we're only allowed to connect to five virtual machines remotely on the cloud, so we 32 00:02:00,540 --> 00:02:04,800 so we don't abuse the resources that collab offers us so nicely for free. 33 00:02:05,580 --> 00:02:11,610 So we've downloaded all images and imported our libraries, so I just rehash what the types of methods 34 00:02:11,610 --> 00:02:12,240 here, huh? 35 00:02:12,270 --> 00:02:18,750 This blog go on Chadwick sites and the site here by Chadwick Todrick. 36 00:02:18,750 --> 00:02:22,860 I should say he has some nice examples of the different resizing methods. 37 00:02:22,860 --> 00:02:23,820 Here you can take a look. 38 00:02:24,270 --> 00:02:26,550 This is basically the input image here. 39 00:02:26,730 --> 00:02:27,900 This is Zoom in a bit. 40 00:02:28,440 --> 00:02:28,890 All right. 41 00:02:29,860 --> 00:02:32,670 Um, so this is a default image, 50 by 50. 42 00:02:33,030 --> 00:02:36,630 And these are the results by reducing the size to 15 by 15. 43 00:02:36,930 --> 00:02:38,580 So you're shrinking the size. 44 00:02:39,000 --> 00:02:41,700 So we're reducing the quality of only 15 by 15. 45 00:02:42,180 --> 00:02:45,510 But we're using a few different interpolation methods to generate this here. 46 00:02:45,600 --> 00:02:54,720 So this is the entire area, how it looks into cubic, into Lonzo's and into nearest and into linear. 47 00:02:55,230 --> 00:03:00,810 So you can see, I think, the best ones, probably into Quebec and into Nereus in this case here into 48 00:03:00,810 --> 00:03:02,670 Lonzo's is quite good as well. 49 00:03:03,660 --> 00:03:09,830 So these are of examples of these different interpolation methods being used here, and you can see 50 00:03:09,840 --> 00:03:13,470 even enlarging the images here when it's actually had a fairly good quality already. 51 00:03:13,950 --> 00:03:17,580 So we enlarge that image now to 150 to 50 to 100. 52 00:03:17,700 --> 00:03:19,860 And you can see what looks, how it looks here. 53 00:03:20,610 --> 00:03:23,460 So what is interpolation actually doing? 54 00:03:23,700 --> 00:03:27,390 So I didn't actually explain it well, actually didn't explain it at all. 55 00:03:27,960 --> 00:03:33,960 Interpolation is basically how is an algorithm to find a value between two points? 56 00:03:34,170 --> 00:03:35,900 Let's take a look at this slide here. 57 00:03:35,940 --> 00:03:40,740 Just briefly bringing this up in during the lesson that can actually put some of this can put this picture 58 00:03:40,740 --> 00:03:43,800 into the lesson that I place in the book for you guys afterward. 59 00:03:44,550 --> 00:03:47,520 So you can see, let's take a look at what interpolation actually is. 60 00:03:47,880 --> 00:03:53,130 So imagine we have a GPS that's logging coordinates as we move along this road of here. 61 00:03:53,610 --> 00:03:56,730 So you can see we took this corner at this point here. 62 00:03:56,760 --> 00:03:59,070 However, we didn't log this green point here. 63 00:03:59,460 --> 00:04:00,990 This is where the actual path was. 64 00:04:01,500 --> 00:04:07,830 So interpolation is going to guess a point in between these two red points that the yellow yellows are 65 00:04:07,830 --> 00:04:10,380 basically the guesses between the points as it moves along. 66 00:04:10,830 --> 00:04:17,460 So it's a way of actually adding more data to existing data to interpolated means as to connect the 67 00:04:17,460 --> 00:04:18,480 ones in between. 68 00:04:19,470 --> 00:04:23,980 So now going back to our lesson here, that's basically what interpolation methods do. 69 00:04:23,980 --> 00:04:29,310 There is when we resize an image, when we enlarge the image, we now have two kind of guests the pixels 70 00:04:29,310 --> 00:04:31,440 that are going to be there that aren't there right now. 71 00:04:31,890 --> 00:04:33,590 Remember, we way we're trying to make. 72 00:04:33,660 --> 00:04:34,710 Let's go back to this blog. 73 00:04:35,010 --> 00:04:37,830 We're trying to make enlarge this image here. 74 00:04:38,400 --> 00:04:43,200 Actually, this is this is the original image here and trying to guess the points it's going to be take 75 00:04:43,200 --> 00:04:44,400 in a new dimension. 76 00:04:45,060 --> 00:04:50,190 So the quality that it's being done is quite important, so into cubic and into these methods. 77 00:04:50,610 --> 00:04:57,810 These are these different formulas that use to apply to actually interpolate and guess the values that 78 00:04:57,810 --> 00:04:58,500 it should be there. 79 00:04:58,740 --> 00:04:59,760 When I say guess, I mean. 80 00:05:00,150 --> 00:05:07,200 Algorithmically, guess the best guess so we can actually get so some in some ways basically looks at 81 00:05:07,200 --> 00:05:13,830 averaging over different pixels in four different areas that's put into cubic does nearest basically 82 00:05:13,830 --> 00:05:15,780 mirrors into political neighborhood pollution. 83 00:05:16,590 --> 00:05:19,200 I'm not going to explain all of these because they actually can get quite technical. 84 00:05:19,650 --> 00:05:20,590 So what are we going to do? 85 00:05:20,610 --> 00:05:23,250 We're just going to resize images in a few different ways here. 86 00:05:23,700 --> 00:05:26,450 So I like resizing a lot. 87 00:05:26,490 --> 00:05:31,880 It's actually quite a useful operation because when you resize an image, we reduce its size. 88 00:05:31,890 --> 00:05:35,100 You actually can speed up a lot of photo operations afterward. 89 00:05:35,490 --> 00:05:39,660 So it's quite a good efficiency technique while you do lose some information. 90 00:05:39,990 --> 00:05:45,630 If you if you use a good interpolation algorithm, you should do what you generally lose a lot from 91 00:05:45,630 --> 00:05:46,590 actually scaling down. 92 00:05:47,580 --> 00:05:48,750 So let's take a look at this. 93 00:05:48,760 --> 00:05:53,820 So the first case here, what we're going to do, we're going to use the argument effects and f way 94 00:05:54,330 --> 00:06:00,510 to reduce the size of an image by 75 percent 0.75 off the original image. 95 00:06:01,260 --> 00:06:03,240 What are we going to do in the second one now? 96 00:06:04,140 --> 00:06:08,820 We're going to double it by using effects equal to fly equal to. 97 00:06:10,110 --> 00:06:16,710 And we're going to use the different interpolation methods doing that twice and then what we're going 98 00:06:16,710 --> 00:06:17,070 to do. 99 00:06:17,520 --> 00:06:20,310 We're going to use the dimensional method, which is actually very useful. 100 00:06:20,400 --> 00:06:26,130 Oftentimes, when, especially when, when you want to keep the aspect ratio off an image, the same 101 00:06:26,130 --> 00:06:27,480 is actually not always possible. 102 00:06:28,410 --> 00:06:32,310 Actually, there's possibilities of what to keep the values the same here, but this is actually quite 103 00:06:32,310 --> 00:06:34,570 good if you actually need to fix a size. 104 00:06:35,040 --> 00:06:40,110 Quite like a function requires image to be an exact size, exact specific size, and you can resize 105 00:06:40,110 --> 00:06:40,350 it. 106 00:06:40,470 --> 00:06:44,640 You can resize any image that size by using this this method here. 107 00:06:45,240 --> 00:06:50,790 So you can just run this here, and we just generally going to resize so images a few times. 108 00:06:52,170 --> 00:06:54,480 So this is with linear interpolation here. 109 00:06:55,320 --> 00:07:00,210 This is the point seventy five scaling right here from the original image. 110 00:07:00,660 --> 00:07:05,700 You can see it doesn't really lose much information, but you can see, though, even though it looks 111 00:07:05,700 --> 00:07:07,110 and looks like the same size. 112 00:07:07,530 --> 00:07:10,680 The pixels go up to 960 here, but now it goes up to 720. 113 00:07:11,370 --> 00:07:17,190 And if you look carefully at a sign this, this zoom into the sign a bit, you can see it's probably 114 00:07:17,460 --> 00:07:21,710 depends on the internet connection and how good your resolution is at your streaming lesson. 115 00:07:22,320 --> 00:07:29,940 But you can see that it actually is lower quality here, especially these words and this less this line 116 00:07:29,940 --> 00:07:32,100 right here compared to this one. 117 00:07:32,820 --> 00:07:35,100 So that's what the resizing did. 118 00:07:35,940 --> 00:07:36,690 Now what are we going to do? 119 00:07:36,720 --> 00:07:38,040 We're going to double the size now. 120 00:07:39,060 --> 00:07:44,760 Now we shouldn't actually get any information mode of it to just look better and bigger when it's printed. 121 00:07:45,030 --> 00:07:53,740 Perhaps that's a good way to a good reason to resize images for printing purposes and scaling the area 122 00:07:53,760 --> 00:07:54,600 can see when we scale. 123 00:07:54,600 --> 00:07:57,150 It's that weird size it actually affected aspect ratio. 124 00:07:57,600 --> 00:08:02,730 So depending on what you need to do, just remember that when you resize to a specific dimension. 125 00:08:03,570 --> 00:08:06,840 So, no, let's move on to using image pyramid's image. 126 00:08:06,840 --> 00:08:12,870 Pyramids are much quicker operation, and you're a quick way to actually scale up a scaled down image. 127 00:08:12,960 --> 00:08:19,470 Basically, double or half the size so you can see this PR people are down. 128 00:08:19,470 --> 00:08:23,490 Pyramid Down takes the image and reduced it to half its size. 129 00:08:23,940 --> 00:08:24,720 Then it actually. 130 00:08:24,720 --> 00:08:28,380 Then we can actually take the same smaller image input and give it back up here. 131 00:08:28,770 --> 00:08:32,730 That would probably give us a pixelated image, which you will see below shortly. 132 00:08:33,150 --> 00:08:37,500 So let's run this function here and see what what we have then. 133 00:08:39,000 --> 00:08:40,680 So let's take a look at this. 134 00:08:40,680 --> 00:08:42,090 So we have the original image here. 135 00:08:42,930 --> 00:08:48,150 We scaled it down here, half the size, so you can see that's half the dimensions of snow 450. 136 00:08:49,290 --> 00:08:52,560 And no, this is the one that was scaled back up. 137 00:08:53,100 --> 00:08:56,340 So we took a smaller image here and then made it bigger again. 138 00:08:56,880 --> 00:08:59,010 And you can see it does lose a lot of information. 139 00:08:59,070 --> 00:08:59,850 It doesn't look. 140 00:09:00,060 --> 00:09:03,270 It looks a bit blurry compared to the original image. 141 00:09:03,400 --> 00:09:10,920 You can see that the details, like in the little walls and windows, is a bit blurry or no, not significantly 142 00:09:10,920 --> 00:09:11,190 so. 143 00:09:11,970 --> 00:09:13,220 And we can do it even more. 144 00:09:13,620 --> 00:09:15,210 But I didn't actually go through this lens. 145 00:09:15,630 --> 00:09:18,090 But what we did here, we actually went. 146 00:09:18,090 --> 00:09:22,330 It went even smaller by actually shrinking it down, shrinking out a smaller image. 147 00:09:22,330 --> 00:09:28,020 So this is Courtnall quarter size of the original image, and you can see it's quite blurry right now. 148 00:09:28,530 --> 00:09:33,420 It's a 200 pixels by three something pixels to something extra something. 149 00:09:34,170 --> 00:09:36,150 So now let's take a look at cropping. 150 00:09:36,150 --> 00:09:42,990 Cropping is a very, very useful technique in computer vision, especially with object detectors or 151 00:09:43,020 --> 00:09:46,710 OCR, where you have to crop segments out of an image. 152 00:09:47,310 --> 00:09:53,790 So how it is cropping with cropping is basically done just actually by using them, by slicing. 153 00:09:54,120 --> 00:09:54,450 OK. 154 00:09:55,710 --> 00:09:56,550 Arrear slicing. 155 00:09:57,030 --> 00:09:59,430 So basically slicing when it is slicing, it means that it is. 156 00:09:59,840 --> 00:10:05,360 Ticking and ticking sections of the era, extracting sections of the array list out. 157 00:10:05,990 --> 00:10:07,190 So how do we do that? 158 00:10:07,340 --> 00:10:08,290 We can actually do it. 159 00:10:08,300 --> 00:10:10,160 Let's quickly go through this year. 160 00:10:10,700 --> 00:10:14,420 So what we're doing here, we're going to heighten wit here. 161 00:10:14,450 --> 00:10:19,250 And the only reason we're getting to Python, which here is because we want to get to start row and 162 00:10:19,250 --> 00:10:24,190 start column here for the cropping that we're going to do so to cropping. 163 00:10:24,200 --> 00:10:25,460 I'll explain cropping shortly. 164 00:10:26,060 --> 00:10:30,700 The section of the middle of the image out from point to five point twenty five two point seventy five 165 00:10:30,710 --> 00:10:31,830 one seven five of the image. 166 00:10:31,830 --> 00:10:36,710 So a quarter basically decent centre crop is what we're trying to do right now. 167 00:10:37,250 --> 00:10:38,290 So this is a line. 168 00:10:38,330 --> 00:10:41,270 Yeah, I need you to pay attention to this line right here. 169 00:10:41,900 --> 00:10:48,620 So this line tells us this is the image here, the input image we're doing, and we're using index indexing 170 00:10:48,620 --> 00:10:53,960 functions, indexing abilities of way to basically get that part of out of out of it. 171 00:10:54,110 --> 00:10:58,220 So we have start row here, which is going to be the height of 0.05. 172 00:10:58,700 --> 00:11:01,340 The end rule, which is the high times, went seven five. 173 00:11:01,730 --> 00:11:03,740 So we're cropping that section here. 174 00:11:03,950 --> 00:11:05,270 And then similarly here. 175 00:11:06,020 --> 00:11:07,790 So I just call these variables. 176 00:11:07,880 --> 00:11:13,670 This here, this actually can be actual numbers if you wanted to put in from the value from the image 177 00:11:13,670 --> 00:11:13,880 here. 178 00:11:14,480 --> 00:11:19,760 So let's just run this now and what we're going to do after we crop the image out. 179 00:11:20,230 --> 00:11:20,930 Call it cropped. 180 00:11:21,680 --> 00:11:28,040 We're going to take that cropped image here, and we're going to actually draw a rectangle around this 181 00:11:28,040 --> 00:11:31,100 onto crop just to basically show the area we cropped out. 182 00:11:34,830 --> 00:11:35,850 So this is it. 183 00:11:36,540 --> 00:11:36,930 So. 184 00:11:38,790 --> 00:11:39,330 Let's take a look. 185 00:11:39,720 --> 00:11:40,710 Sorry for all the screwing. 186 00:11:41,880 --> 00:11:43,320 This is the original image here. 187 00:11:43,710 --> 00:11:45,690 And remember I said, we're going to take a rectangle. 188 00:11:46,140 --> 00:11:50,640 The rectangle basically is just throwing a yellow block over the area we cropped out. 189 00:11:50,730 --> 00:11:51,930 So this is your of away. 190 00:11:51,930 --> 00:11:56,250 So I wanted to go one two five down two point two seven point seventy five down. 191 00:11:56,820 --> 00:11:57,820 So let's take a look. 192 00:11:57,840 --> 00:12:02,850 That's exactly what we did point to five here, two point seventy five here of the image and in terms 193 00:12:02,850 --> 00:12:08,580 of ratios from the when the width and height and we were just displaying it. 194 00:12:08,590 --> 00:12:11,100 Now this is the full crop here. 195 00:12:11,130 --> 00:12:11,790 We dropped out. 196 00:12:12,510 --> 00:12:13,380 So that's it. 197 00:12:13,380 --> 00:12:14,340 That's it for cropping. 198 00:12:14,370 --> 00:12:15,660 Hope you enjoyed this lesson. 199 00:12:16,140 --> 00:12:20,460 In the next section, we'll take a look at arithmetic and Bitwise operations. 200 00:12:20,730 --> 00:12:21,960 So stay tuned for that. 201 00:12:22,020 --> 00:12:22,470 Thank you.