1 00:00:01,530 --> 00:00:09,030 The input to our network when we are doing image recognition is an image. 2 00:00:09,030 --> 00:00:16,560 Now that image can be either black and white which is also known as grayscale image or it can be colored 3 00:00:18,480 --> 00:00:20,490 in a grayscale image. 4 00:00:20,610 --> 00:00:29,890 Each pixel contains only one information and the information is how wide is that pixel. 5 00:00:30,270 --> 00:00:38,880 So usually it is seen on a scale of 0 to 255 255 means completely white. 6 00:00:38,940 --> 00:00:46,340 Zero means completely black and any value in between 0 and 255 is a shade of gray. 7 00:00:49,110 --> 00:00:54,930 But in a coloured image each pixel has three values instead of one. 8 00:00:57,030 --> 00:01:05,910 These three are R G and B that is red green and blue. 9 00:01:06,000 --> 00:01:13,890 It turns out that receptors in human eye perceive color as the combination of these three colors only 10 00:01:15,750 --> 00:01:18,290 these are called primary colors. 11 00:01:19,530 --> 00:01:24,330 If you mix these three in different quantities you will get different colors 12 00:01:27,510 --> 00:01:39,630 so each pixel has three values value for each of these primary color varies from 0 to 255 and depending 13 00:01:39,630 --> 00:01:42,890 on what combination of these three values do we have. 14 00:01:43,620 --> 00:01:46,160 We can see the corresponding color in the pixel. 15 00:01:48,300 --> 00:01:53,150 So if all three values are 255 we get light color. 16 00:01:55,350 --> 00:02:00,780 If red and green are 255 and blue is 0 we get yellow 17 00:02:04,440 --> 00:02:06,640 hair in this animation. 18 00:02:06,900 --> 00:02:16,440 You can see that when all three colors overlap we get white color when red and green overlap we get 19 00:02:16,440 --> 00:02:20,160 yellow when blue and green overlap. 20 00:02:20,160 --> 00:02:25,370 We get to see on and on here. 21 00:02:25,410 --> 00:02:26,970 I have also mentioned one example. 22 00:02:27,840 --> 00:02:35,430 If you want to create but particular here are the values of our G and B setting these values will give 23 00:02:35,430 --> 00:02:38,910 you by particular. 24 00:02:39,570 --> 00:02:48,870 So if each pixel has three values we can say that the entire image is made up of three layers one layer 25 00:02:49,260 --> 00:02:53,060 made up of red color values called the red channel. 26 00:02:53,760 --> 00:03:01,260 One layer made above green color values called green channel and one layer made up of blue color values 27 00:03:01,470 --> 00:03:02,810 called the Blue channel. 28 00:03:04,350 --> 00:03:12,840 So you need to remember that whenever we work with a grayscale dataset such as fashion amnesty you have 29 00:03:12,930 --> 00:03:14,230 only one channel. 30 00:03:15,180 --> 00:03:20,990 But whenever we work on color image we need to take into account the three color channels. 31 00:03:21,000 --> 00:03:21,390 Also 32 00:03:24,840 --> 00:03:30,480 to demonstrate to you a color image is made up of three channels. 33 00:03:30,480 --> 00:03:34,740 I have created pixels using cells of Excel worksheet 34 00:03:37,370 --> 00:03:42,280 contended that this pixel is a bundle of these three cells. 35 00:03:42,510 --> 00:03:52,830 So the first pixel is bundle of three cells one cell to take rate value one where a green and one will 36 00:03:52,830 --> 00:04:04,290 take Blue value and the way that they make one pixel the value in these cells correspond to R G and 37 00:04:04,290 --> 00:04:10,320 B values and we have used conditional formatting to color these cells depending on their value 38 00:04:16,320 --> 00:04:24,420 so if you look at this cell this cell has value to thirty one so it is giving us a bright green color 39 00:04:26,190 --> 00:04:35,120 the cell to the right of this has only forty seven value which is why it is dark so if it is zero it 40 00:04:35,120 --> 00:04:36,660 will be completely black. 41 00:04:36,660 --> 00:04:46,980 If it is 255 it will be bright green and anything in between is a grayscale of green color it similarly 42 00:04:46,980 --> 00:04:52,580 walks with red and blue general also. 43 00:04:52,650 --> 00:05:01,310 Now if we are looking at individual pixels we cannot make out the image but when we start zooming out 44 00:05:04,710 --> 00:05:09,690 you start seeing the image. 45 00:05:10,030 --> 00:05:13,560 Do you see a normal colored image of the model 46 00:05:16,240 --> 00:05:17,500 isn't this somewhat magical. 47 00:05:18,550 --> 00:05:24,230 We were able to create a colored image using only three colors. 48 00:05:24,460 --> 00:05:34,930 You can see on the borders other than the model you can see white color although there were only red 49 00:05:35,050 --> 00:05:37,530 green and blue colors. 50 00:05:37,550 --> 00:05:46,010 This is because our eyes perceive a combination of these three colors as white. 51 00:05:46,030 --> 00:05:53,800 This is our computer and mobile phone screens also look for each point or pixel on the screen. 52 00:05:53,890 --> 00:05:59,980 There are three small light bulbs one for each red green and blue color. 53 00:06:01,450 --> 00:06:11,810 And with these three channels only we see so many beautifully colored images on our screens. 54 00:06:11,910 --> 00:06:13,310 I would like to mention here. 55 00:06:13,350 --> 00:06:17,740 This website which I used for creating this Excel image thing. 56 00:06:17,890 --> 00:06:26,400 Matt Dore Gold or UK you can just upload any image here and it creates this are G.B. pixilated excel 57 00:06:26,400 --> 00:06:27,770 file for it. 58 00:06:28,170 --> 00:06:29,960 It is an amazing and fun tool. 59 00:06:29,970 --> 00:06:30,690 Do check it out.