1 00:00:00,060 --> 00:00:04,070 So now we move on to Section 21, which is filtering colors. 2 00:00:04,110 --> 00:00:06,360 So go ahead and open this notebook. 3 00:00:06,870 --> 00:00:11,370 So if this thing is seen, this input, which I've highlighted here, is this HSV color space. 4 00:00:11,910 --> 00:00:16,950 Now, I don't know if you recall from our earlier lesson where we talked about the HSV color space stands 5 00:00:16,950 --> 00:00:18,940 for Hue saturation and the value. 6 00:00:19,590 --> 00:00:24,540 Basically, this allows us to this this color space allows us to filter on colors. 7 00:00:24,840 --> 00:00:28,620 And the reason is because we can actually filter on Hue quite easily. 8 00:00:28,650 --> 00:00:31,590 You can just extract ranges between values. 9 00:00:31,590 --> 00:00:38,520 Here would be it's much, much, much more difficult to actually filter out a specific color, mainly 10 00:00:38,520 --> 00:00:42,080 because different combinations of RGV give different colors. 11 00:00:42,090 --> 00:00:43,590 So it's I mean, you can do it. 12 00:00:43,620 --> 00:00:46,230 It's just much harder than HSV makes it much easier. 13 00:00:46,920 --> 00:00:48,420 So let's move on. 14 00:00:48,690 --> 00:00:55,410 So firstly, let's load all images and libraries and create a major function, which I've done already. 15 00:00:55,470 --> 00:00:57,960 You can go ahead and run this block of code. 16 00:00:58,620 --> 00:01:06,120 So what we're going to do, we're going to actually create a filter that basically removes the truck 17 00:01:06,120 --> 00:01:09,030 and this land here and only keeps the sky. 18 00:01:09,060 --> 00:01:11,160 So we're making a blue or purple filter. 19 00:01:11,640 --> 00:01:15,660 So to do that, you can see we need to defined an upper and lower range. 20 00:01:16,170 --> 00:01:20,580 So this upper and lower was this one up, I should say, started at 90. 21 00:01:20,610 --> 00:01:21,750 Remember, this is each. 22 00:01:22,020 --> 00:01:24,990 So we go from 90 to 125, that's all orange. 23 00:01:25,590 --> 00:01:29,910 And then we go zero to 255 30 to 50 540 saturation in value. 24 00:01:29,910 --> 00:01:32,520 Since we we don't we don't care about saturation and value. 25 00:01:32,520 --> 00:01:35,580 Even though we can, we can make filters for this. 26 00:01:36,120 --> 00:01:38,550 In this case, we just want to get all the ranges of the blue. 27 00:01:39,060 --> 00:01:43,450 So 90 to 135, let's go back up to a color wheel 90. 28 00:01:44,310 --> 00:01:49,800 You can see is where the blue starts around here and one two five is red and basically here. 29 00:01:49,920 --> 00:01:54,290 So even though I did call the purple, it's actually a blue filter. 30 00:01:54,300 --> 00:01:55,800 Not then why I left purple then. 31 00:01:56,460 --> 00:02:03,120 So let's run this block of code and we can see the outputs and I'll go back and you can see it's exactly 32 00:02:03,120 --> 00:02:03,750 what we wanted. 33 00:02:04,140 --> 00:02:10,050 Now you can see there's an interim image here, which is a banner raised image that looks a bit like 34 00:02:10,050 --> 00:02:10,620 a mask. 35 00:02:10,980 --> 00:02:12,910 And that's because it is a mask. 36 00:02:13,320 --> 00:02:14,580 So let's go over to code here. 37 00:02:14,580 --> 00:02:19,290 So what we did first thing, we loaded the image, we defined our ranges and then we converted to the 38 00:02:19,290 --> 00:02:21,210 color image to an HSV image. 39 00:02:21,690 --> 00:02:27,270 And then we use the C v2.0 in range of function with the upper limit and lower image here. 40 00:02:27,390 --> 00:02:30,120 Sorry, and the image on the upper, the lower and upper limits. 41 00:02:30,720 --> 00:02:31,920 And that gives us a mess. 42 00:02:32,820 --> 00:02:38,820 The mask is a binary threshold at image, which basically tells us gives us everything. 43 00:02:38,820 --> 00:02:42,810 That's everything that's in the filter is white on it. 44 00:02:42,840 --> 00:02:48,100 So it's a yes, and everything that's not in the filter range is black. 45 00:02:48,150 --> 00:02:48,670 So that's like a. 46 00:02:49,560 --> 00:02:52,860 And do you remember our Bitwise operations? 47 00:02:53,280 --> 00:03:00,030 Well, we can use a Bitwise operation and which basically tells us that if we do use and with this image 48 00:03:00,480 --> 00:03:06,600 and our mask and the statistics and image twice, you know, if you recall in a previous lesson, we 49 00:03:06,600 --> 00:03:09,750 can actually create this nice filter here. 50 00:03:10,620 --> 00:03:14,730 So you can see we've extracted everything except the blue sky, which is pretty cool. 51 00:03:15,660 --> 00:03:20,040 Filtering is actually really good if you want to isolate, see on like a tennis ball, which you know 52 00:03:20,040 --> 00:03:22,920 is a specific color or cricket ball, which would be red or white. 53 00:03:23,580 --> 00:03:25,140 So all basketball would you be orange? 54 00:03:25,560 --> 00:03:27,720 So you can do the filters and then do tracking like that. 55 00:03:27,720 --> 00:03:28,590 It works quite well. 56 00:03:29,190 --> 00:03:30,750 So let's do this. 57 00:03:30,750 --> 00:03:33,300 Let's actually look up that color wheel. 58 00:03:34,140 --> 00:03:35,700 What if we wanted to filter on the red? 59 00:03:36,360 --> 00:03:39,420 You can see red is a bit tricky because red is at the threshold down here. 60 00:03:39,420 --> 00:03:43,920 Right here, red actually goes from zero to maybe around 15. 61 00:03:44,550 --> 00:03:48,960 And then all the way back down to this is this is a 180 here. 62 00:03:48,990 --> 00:03:50,970 So 180 to maybe like 160. 63 00:03:51,540 --> 00:03:57,270 But that's a problem because now we have two oranges, we have zero to 15 and then one, 65 to 180. 64 00:03:57,690 --> 00:04:00,000 So how do we do it with two color ranges? 65 00:04:00,600 --> 00:04:05,930 Well, it's it's actually not that hot, but we do have to specify to mess. 66 00:04:06,420 --> 00:04:11,970 So you create a mask, the a mask, which is the zero to 10, and then we create other masks, which 67 00:04:11,970 --> 00:04:17,370 is 170 to 180, and we create the mask is Mesirow Mask one. 68 00:04:18,030 --> 00:04:23,700 So then we joined the mask together by adding them, and then we can do the same Bitwise operation again. 69 00:04:24,120 --> 00:04:28,110 So in this image with Hillary Clinton, we're going to filter out the Reds in the image. 70 00:04:28,620 --> 00:04:30,270 So let's run this code. 71 00:04:33,150 --> 00:04:34,410 Images and find. 72 00:04:38,110 --> 00:04:42,520 Oh, that's because this variable needs to be image here. 73 00:04:42,700 --> 00:04:45,370 So sorry for my mistake. 74 00:04:46,510 --> 00:04:47,550 OK, so there we go. 75 00:04:47,600 --> 00:04:53,290 So it works now, so you can see this is a mess that we added together so we can actually get everything 76 00:04:53,290 --> 00:04:53,890 that's ready. 77 00:04:53,890 --> 00:04:58,940 Or at least reddish tint skin tones apparently were red in this image and actually do tell it this bit 78 00:04:58,960 --> 00:04:59,220 red. 79 00:05:00,370 --> 00:05:03,640 That's why it was in that color range, and that's left. 80 00:05:03,820 --> 00:05:09,640 So we get scary vision of Hillary Clinton with missing teeth and missing here. 81 00:05:10,300 --> 00:05:10,930 But it's OK. 82 00:05:11,170 --> 00:05:11,890 Sorry, Hillary. 83 00:05:12,610 --> 00:05:15,580 And we have her red dress filtered, so that's pretty cool. 84 00:05:16,060 --> 00:05:18,390 So that concludes this lesson and color filtering. 85 00:05:18,400 --> 00:05:22,870 It's a very useful lesson for many different applications, so it's a good thing. 86 00:05:22,870 --> 00:05:24,150 It's named filtering colors. 87 00:05:24,160 --> 00:05:24,880 It's easy to find. 88 00:05:25,390 --> 00:05:31,810 So we will now move on to something called the watershed algorithm, which is a marker based image segmentation 89 00:05:31,810 --> 00:05:32,350 algorithm. 90 00:05:32,620 --> 00:05:34,780 So stay tuned and I'll see you in the next lesson. 91 00:05:35,020 --> 00:05:35,440 Thank you.