WEBVTT 1 00:00:01.000 --> 00:00:04.000 For our last sneak, I'll first ask have you ever used Photoshop, Rainn? 2 00:00:04.000 --> 00:00:06.000 Yes. Yes, I have. 3 00:00:06.000 --> 00:00:11.000 Are you in the audience ready for the annual tradition of the "mind-blowing" Photoshop sneak? 4 00:00:11.000 --> 00:00:12.000 Yeah! 5 00:00:12.000 --> 00:00:15.000 Well, you've come to the right place. 6 00:00:15.000 --> 00:00:17.000 I'm not even going to set this one up. 7 00:00:17.000 --> 00:00:21.000 I'm just going to bring out Jue Wang to show you a sneak we have around Photoshop 8 00:00:21.000 --> 00:00:23.000 that I think will just blow your mind. 9 00:00:23.000 --> 00:00:25.000 Can I get a free Photoshop out of this? 10 00:00:25.000 --> 00:00:29.000 I'll tell you what--if you'll sign one of the Rainn Wilson bobbleheads 11 00:00:29.000 --> 00:00:34.000 for Teresa, who is one of our admins at Adobe, I'll get you a copy of Photoshop. 12 00:00:34.000 --> 00:00:38.000 Really? >>Yes. >>Sweet! Done. >>Done. 13 00:00:39.000 --> 00:00:41.000 That's Teresa. >>Hi, Teresa. 14 00:00:41.000 --> 00:00:45.000 Let's bring out Jue Wang and let's see some cool Photoshop magic. Jue? 15 00:00:57.000 --> 00:01:04.000 Hello, everyone. Here you see a crappy, blurry image that one of my friends took. 16 00:01:04.000 --> 00:01:08.000 He was probably drunk that day like many of your right now. 17 00:01:08.000 --> 00:01:11.000 He couldn't hold his camera steady when he shot this one. 18 00:01:11.000 --> 00:01:15.000 He knows I work at Adobe, so he sent this one to me to see 19 00:01:15.000 --> 00:01:19.000 if I can do something special to remove the blur and to restore a sharp image. 20 00:01:19.000 --> 00:01:23.000 It turns out he was lucky, because we have been working on this problem for a while, 21 00:01:23.000 --> 00:01:26.000 and we have built a prototype to solve it. 22 00:01:26.000 --> 00:01:33.000 To show it, I will just load the image into the Photoshop plugin we developed. 23 00:01:33.000 --> 00:01:39.000 You can see there aren't a lot of controls here, because it's still an early prototype. 24 00:01:39.000 --> 00:01:46.000 To save some time, I'll just load some predefined parameters, which contain only a few numbers. 25 00:01:46.000 --> 00:01:51.000 The first thing to do with this tool is to click the "analyze" button, 26 00:01:51.000 --> 00:01:54.000 which analyses how the image was blurred in the first place. 27 00:01:54.000 --> 00:01:57.000 The system does the heavy lifting under the hood, 28 00:01:57.000 --> 00:02:03.000 which involves some intensive computation that I won't try to explain here. 29 00:02:03.000 --> 00:02:07.000 The output of this step is something we call a "blur kernel," 30 00:02:07.000 --> 00:02:12.000 which you will see in a moment--this is a kind of slow computer. Ah, now here it is. 31 00:02:12.000 --> 00:02:18.000 It's essentially a grayscale image showing how the image was blurred. 32 00:02:18.000 --> 00:02:20.000 Formally, it's called point-spread function, 33 00:02:20.000 --> 00:02:26.000 but you can also view it as the motion trajectory of the camera when the shutter was open. 34 00:02:26.000 --> 00:02:33.000 The next thing to do is to use it to restore a sharp image. 35 00:02:33.000 --> 00:02:36.000 And here it is. 36 00:02:36.000 --> 00:02:44.000 No way! >>Yes. >>No way. That's impossible! >>Yes. >>That's two photographs. >>Nope. 37 00:02:44.000 --> 00:02:50.000 Jue, there's no way you did that. >>Yes way, Jue. 38 00:02:50.000 --> 00:02:54.000 No way, Jue. 39 00:02:54.000 --> 00:02:57.000 So you can see it's pretty sharp here. 40 00:02:57.000 --> 00:03:01.000 How does it do that? How do you do that? 41 00:03:01.000 --> 00:03:04.000 There is an algorithm behind it. 42 00:03:04.000 --> 00:03:06.000 A what? >>An algorithm behind it. 43 00:03:06.000 --> 00:03:10.000 Algorithm? What's an algorithm? 44 00:03:10.000 --> 00:03:14.000 We'll talk at Weezer. >>Okay. 45 00:03:14.000 --> 00:03:18.000 If you look at this small poster--I'm going to switch back to the original-- 46 00:03:18.000 --> 00:03:22.000 you can see how much improvement we have done on this example. 47 00:03:22.000 --> 00:03:26.000 Now, our system not only makes blurry images look better, 48 00:03:26.000 --> 00:03:30.000 it can also help recover important information from the image. 49 00:03:30.000 --> 00:03:34.000 Here is a poster image I shot using my cell phone camera. 50 00:03:34.000 --> 00:03:38.000 My thought was to record a poster so I can read it at home, 51 00:03:38.000 --> 00:03:41.000 only to find out it is too blurry so I cannot read anything. 52 00:03:41.000 --> 00:03:44.000 Let's see if we can improve this image. 53 00:03:44.000 --> 00:03:51.000 I just load the image into the plugin. For now, I load some pre-selected parameters. 54 00:03:51.000 --> 00:03:56.000 Because this is a relatively large image, I'm not going to use the whole image for analysis, 55 00:03:56.000 --> 00:03:59.000 which will take some time on this computer. 56 00:03:59.000 --> 00:04:09.000 I'm just going to select a local region and ask the system to only analyze this local region. 57 00:04:09.000 --> 00:04:13.000 This is a special case, because the image contains text. 58 00:04:13.000 --> 00:04:16.000 We have to tell the system that we are dealing with text 59 00:04:16.000 --> 00:04:22.000 so the system knows to apply some special treatment 60 00:04:22.000 --> 00:04:25.000 Would this be in the next version of Photoshop, for real? 61 00:04:25.000 --> 00:04:30.000 Remember we said "sneaks." They may or may not appear in a future version. 62 00:04:30.000 --> 00:04:34.000 We don't commit to timing for anything in a sneak. 63 00:04:34.000 --> 00:04:37.000 You should definitely do this. 64 00:04:37.000 --> 00:04:43.000 We can see the blur kernel here, and similarly we can define another preview window, 65 00:04:43.000 --> 00:04:51.000 which allows you to preview local results to see if it is good. Here it is. >>Wow! 66 00:04:51.000 --> 00:04:56.000 You can see that you can read the phone number. 67 00:04:56.000 --> 00:05:00.000 Who's the boss here? Who's in charge of all this? Oh it was you. 68 00:05:00.000 --> 00:05:05.000 You two guys--you should do this in the next Photoshop. 69 00:05:05.000 --> 00:05:11.000 People will really--seriously, I'm just a chump--people will really like this. 70 00:05:11.000 --> 00:05:14.000 People will love it. 71 00:05:14.000 --> 00:05:19.000 Would I pay for it? No, absolutely not. I don't pay for anything any more. 72 00:05:19.000 --> 00:05:25.000 I'm a minor television celebrity, so I get free shit. 73 00:05:27.000 --> 00:05:31.000 All right. Just one more comparison before and after. 74 00:05:31.000 --> 00:05:34.000 I have one last quick demo. 75 00:05:34.000 --> 00:05:39.000 This is Kevin Lynch, our CEO, at last year's MAX. 76 00:05:39.000 --> 00:05:44.000 We found this image online. It's a random internet photo. It's bad. 77 00:05:44.000 --> 00:05:48.000 Let's see if we can improve it. 78 00:05:48.000 --> 00:05:57.000 Because this is a relatively small image--I still need to load some parameters-- 79 00:05:57.000 --> 00:06:06.000 I'm just going to analyze this image, which should be pretty fast, and restore it. 80 00:06:06.000 --> 00:06:11.000 You can see it's much better. >>Wow. 81 00:06:11.000 --> 00:06:14.000 This is image deblurring. Thank you very much. 82 00:06:14.000 --> 00:06:17.000 Thank you, Jue. >>Wow. 83 00:06:17.000 --> 00:06:25.000 Pretty amazing. >>So every year you get a Photoshop thing that just blows your mind? 84 00:06:25.000 --> 00:06:28.000 We have a history of mind-blowing Photoshop demos at Sneaks. 85 00:06:28.000 --> 00:06:32.000 A couple years ago you would've seen things like automatically removing 86 00:06:32.000 --> 00:06:36.000 elements from pictures and seamlessly correcting the background, 87 00:06:36.000 --> 00:06:39.000 which actually did ship in Photoshop CS5. 88 00:06:39.000 --> 00:06:43.000 That's an example of something we showed at a Sneak that's in shipping software today. 89 00:06:43.000 --> 00:06:46.000 Yes, exactly. >>Hey, that's incredible.