1 00:00:05,480 --> 00:00:06,680 Welcome back. 2 00:00:06,710 --> 00:00:11,480 Now that we understand how to work with the basics of dumpsite arrays Let's quickly talk about what 3 00:00:11,480 --> 00:00:14,140 an image actually is. 4 00:00:14,140 --> 00:00:19,450 So before we talk about how to use pi specifically if image files Let's first discuss how computers 5 00:00:19,450 --> 00:00:21,460 themselves handle images. 6 00:00:21,460 --> 00:00:26,620 Let's imagine that we want to build software that could sort out mail based on the zip code of the address 7 00:00:26,920 --> 00:00:33,020 written on the envelope often mail is handwritten which means we would somehow need to begin to understand 8 00:00:33,290 --> 00:00:38,100 how to use computer vision to actually read in image data of these handwritten digits. 9 00:00:38,120 --> 00:00:43,130 So most likely we're going to be working with an image such as a photograph of the envelope like a TNG 10 00:00:43,160 --> 00:00:43,930 or jpeg. 11 00:00:44,000 --> 00:00:47,850 And we want to understand how does a computer actually represent image data. 12 00:00:47,870 --> 00:00:50,980 Let's imagine we have a simple image of a handwritten number. 13 00:00:50,990 --> 00:00:54,410 So here we have images of numbers 1 2 and 3. 14 00:00:54,470 --> 00:00:57,100 And these are either jpeg or PNB files. 15 00:00:58,020 --> 00:01:01,770 Each single digit image can be actually represented as an array. 16 00:01:01,800 --> 00:01:07,600 So here we have the numbers zero through 9 and these are all being represented as an array. 17 00:01:07,600 --> 00:01:12,880 So for example here is a number represented by 28 by 28 pixels. 18 00:01:12,880 --> 00:01:17,620 You can imagine that as we get more and more pixels we'll have finer and finer resolution. 19 00:01:17,620 --> 00:01:23,110 You'll also notice that the image on the left is grayscale meaning it's only black or white or some 20 00:01:23,110 --> 00:01:29,220 color of gray and it's being represented as an array with values ranging from 0 to 1. 21 00:01:29,440 --> 00:01:34,550 So how dark a pixel should be can be represented as those values between 0 and 1. 22 00:01:34,600 --> 00:01:40,180 So you'll notice here we have a number one as the hand-written digit and where it's Darcus it's represented 23 00:01:40,270 --> 00:01:42,850 as a single digit one in or it's completely white. 24 00:01:42,880 --> 00:01:44,890 It's represented as 0. 25 00:01:44,890 --> 00:01:52,630 Now often default images when you actually read them in they have values between 0 and 255 where 255 26 00:01:52,690 --> 00:01:56,700 is the highest or brightest value and zero is the darkest. 27 00:01:56,940 --> 00:02:02,910 The range 0 to 255 that actually has to do with how computer store 8 bit numbers that's a bit outside 28 00:02:02,910 --> 00:02:06,990 the scope of this course and we really don't need to understand it but at the end I'll show you a couple 29 00:02:07,020 --> 00:02:09,540 links that you can check out for more information. 30 00:02:10,780 --> 00:02:16,900 Now if you ever wanted to get those values between 0 0 and 255 to be between 0 and 1 all you would need 31 00:02:16,900 --> 00:02:19,430 to do is divide by the max value to 5. 32 00:02:19,600 --> 00:02:22,210 In order to normalize to between 0 and 1. 33 00:02:22,210 --> 00:02:26,000 And later on in the course we'll see some examples of doing that. 34 00:02:26,290 --> 00:02:30,150 So we've understood how grayscale images can be represented as arrays. 35 00:02:30,280 --> 00:02:38,090 But what about color images color images can be represented as a combination of red green and blue additive 36 00:02:38,090 --> 00:02:43,670 color mixing allows us to represent a wide variety of colors by simply combining different amounts of 37 00:02:43,850 --> 00:02:52,160 our GP or red green blue and our GP allows us to produce a range of colors and here we can see the entire 38 00:02:52,160 --> 00:02:58,790 dynamic range that just red green and blue allows you to create this sort of image is known as a color 39 00:02:58,790 --> 00:03:04,640 triangle and Biscoe what that means is only colors within this triangle can be reproduced by mixing 40 00:03:04,640 --> 00:03:10,400 the primary colors that is red green and blue colors outside the color triangle are therefore shown 41 00:03:10,400 --> 00:03:17,480 here is gray and the primaries and the D6 five that point in the middle that's the white point and basically 42 00:03:17,480 --> 00:03:23,450 shows you the edges of what's producible and keep in mind that humans themselves have limits as far 43 00:03:23,450 --> 00:03:29,420 as what colors they can see but obviously red green and blue allows for quite a wide range later on 44 00:03:29,420 --> 00:03:34,130 in the course we're going to be learning about alternative color mappings that can be applied to images 45 00:03:34,250 --> 00:03:36,410 that go beyond just red green and blue. 46 00:03:36,470 --> 00:03:40,600 But for now red green and blue will be more than enough for us. 47 00:03:40,670 --> 00:03:45,530 So each color channel will have intensity values and you have already seen this sort of representation 48 00:03:45,530 --> 00:03:46,480 in other software. 49 00:03:46,500 --> 00:03:51,590 RG backsliders In fact you can just Google R.G. Boxleitner to play around these sort of values. 50 00:03:51,620 --> 00:03:56,300 So here you can play around the valleys of red green and blue to produce different colors all the way 51 00:03:56,510 --> 00:03:58,900 values from 0 to 255. 52 00:04:00,830 --> 00:04:06,700 So what does that actually mean when you've read in a color image with python and a computer in general. 53 00:04:06,980 --> 00:04:11,690 So what happens is when you read in an image it's going to actually have three dimensions a height that 54 00:04:11,720 --> 00:04:18,870 mention it with dimension and then through them actions for these color channels red green and blue. 55 00:04:18,950 --> 00:04:22,330 So this means that you're reading in the image and if you're a check it's shape. 56 00:04:22,370 --> 00:04:23,970 If you read it as an umpire. 57 00:04:24,260 --> 00:04:28,940 The shape is going to be something like 12 A.D. by 720 by three. 58 00:04:28,940 --> 00:04:34,950 So that would be the pixel with the pixel height and then the actual three color channels. 59 00:04:34,970 --> 00:04:40,430 So if we were to actually visualize this basically you have a three dimensional array where we have 60 00:04:40,430 --> 00:04:43,480 the typical to them mentions except now you have three of those. 61 00:04:43,520 --> 00:04:46,560 One for each major color. 62 00:04:46,870 --> 00:04:51,060 Keep in mind the computer itself won't know that a channel is red. 63 00:04:51,130 --> 00:04:56,500 It just knows that there are now three intensity channels essentially three sort of grayscale channels 64 00:04:56,800 --> 00:04:58,730 representing a color. 65 00:04:58,750 --> 00:05:02,950 So notice that when the computer actually reads it in it gets something that looks like this last column 66 00:05:02,950 --> 00:05:04,540 here black and white. 67 00:05:04,540 --> 00:05:09,400 It's up to whatever software are using to actually represent those as color in order to combine them 68 00:05:09,430 --> 00:05:11,700 to get that left image all the way on the left. 69 00:05:11,820 --> 00:05:12,910 Or we have the parrot. 70 00:05:13,090 --> 00:05:16,500 So keep in mind we're still getting values peine 0 and 255. 71 00:05:16,620 --> 00:05:23,000 It's just now we have three of those mappings one for red one for green and one for blue. 72 00:05:23,030 --> 00:05:27,920 So the user needs to dictate which channel is for which color each channel alone is essentially the 73 00:05:27,920 --> 00:05:29,910 same as a single greyscale image. 74 00:05:31,370 --> 00:05:34,130 So let's explore this further if some pie in Python. 75 00:05:34,310 --> 00:05:38,830 I also highly encourage you to check out the Wikipedia article on R.G. color channeling. 76 00:05:38,870 --> 00:05:44,120 It's called RGV color model and has a lot of interesting details as well as the history of our G-B color 77 00:05:44,120 --> 00:05:47,620 model and a couple of more fun things to play around with. 78 00:05:47,630 --> 00:05:48,060 OK. 79 00:05:48,290 --> 00:05:51,290 So let's move on and explore this further with non-high in Python. 80 00:05:51,290 --> 00:05:52,260 I'll see you at the next lecture.