Adobe Previews Photoshop Algorithm to Reverse Severe Camera Shake

Peter Elst

By calculating the camera sensor's motion path during exposure into a point spread function, the prototype restores lost edge contrast to recover smeared photos and unreadable text.

AI-assisted summary of Adobe MAX 2011 - Photoshop Image Deblurring sneak — original source. Check the original for full context.

Source published: 2011-10-07

Key ideas

Adobe previewed an experimental Photoshop deblurring prototype at MAX 2011 that mathematically reverses severe camera shake by calculating the sensor's physical motion path. (0:00)
The tool computes a point spread function (PSF) or grayscale blur kernel, running a deconvolution filter to restore edge contrast and illegible text. (0:00)
Specialized settings allow targeted mathematical processing on designated text regions, expediting analysis via localized bounding boxes. (2:07)
Adobe reiterated its policy of not committing to release schedules for sneak peeks, though previous MAX prototypes like Content-Aware Fill successfully graduated to commercial releases in Photoshop CS5. (2:07)

Detailed notes

## Background Published on YouTube by Peter Elst on October 7, 2011, this video documents an Adobe MAX 2011 sneak peek presentation showcasing an experimental image deblurring technology for Photoshop. The description provides a link directing viewers to a collection of additional sneak peek videos from the conference on Elst's personal blog.

## TL;DR An Adobe researcher presents an early prototype Photoshop plug-in capable of calculating camera movement during shutter exposure to reverse severe motion blur. By computing a point spread function (blur kernel), the tool utilizes deconvolution filters to restore sharp edges and lost detail in photographs. Live tests demonstrate the technology deblurring a camera snapshot, extracting unreadable text from a smeared mobile photo, and sharpening a low-resolution internet image of Adobe CTO Kevin Lynch. Guest co-host Rainn Wilson joins on stage to provide comedic commentary as the hosts contextualize the project alongside previous successful sneaks like Content-Aware Fill.

## Atmosphere & style The presentation carries the electric, slightly unpolished excitement of an early-2010s tech conference sneak peek, balancing dense computational science with freewheeling stage humor. The mood thrives on the playful clash between the presenter's matter-of-fact algorithmic walkthroughs and Rainn Wilson's deadpan skepticism and celebrity riffs. Watching heavily blurred, illegible images snap into crisp focus generates genuine moments of audience astonishment, shifting the room from curious anticipation to spontaneous cheers and applause.

## Bulletin - [00:00](https://www.youtube.com/watch?v=xxjiQoTp864&t=0s) Adobe previewed an experimental Photoshop deblurring prototype at MAX 2011 that mathematically reverses severe camera shake by calculating the sensor's physical motion path. - [00:00](https://www.youtube.com/watch?v=xxjiQoTp864&t=0s) The tool computes a point spread function (PSF) or grayscale blur kernel, running a deconvolution filter to restore edge contrast and illegible text. - [02:07](https://www.youtube.com/watch?v=xxjiQoTp864&t=127s) Specialized settings allow targeted mathematical processing on designated text regions, expediting analysis via localized bounding boxes. - [02:07](https://www.youtube.com/watch?v=xxjiQoTp864&t=127s) Adobe reiterated its policy of not committing to release schedules for sneak peeks, though previous MAX prototypes like Content-Aware Fill successfully graduated to commercial releases in Photoshop CS5.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=xxjiQoTp864&t=0s) The prototype Photoshop plug-in is engineered to isolate camera shake and completely reverse severe motion blur through computational analysis. - [00:00](https://www.youtube.com/watch?v=xxjiQoTp864&t=0s) The calculation determines the camera sensor's trajectory during exposure, generating a grayscale point spread function (PSF) known as a blur kernel. - [00:00](https://www.youtube.com/watch?v=xxjiQoTp864&t=0s) Applying the calculated motion path across the image through deconvolution restores edge contrast and fine visual details that appeared lost. - [00:00](https://www.youtube.com/watch?v=xxjiQoTp864&t=0s) Due to the prototype interface's technical complexity, pre-configured numerical parameters were loaded on stage to initiate the automated heavy analysis. - [00:00](https://www.youtube.com/watch?v=xxjiQoTp864&t=0s) Guest co-host Rainn Wilson reacted with vocal disbelief as the first severely degraded image snapped into sharp focus. - [02:07](https://www.youtube.com/watch?v=xxjiQoTp864&t=127s) Drawing a bounding box around a localized region of interest prevents the system from having to calculate an entire high-resolution frame, dramatically cutting processing time. - [02:07](https://www.youtube.com/watch?v=xxjiQoTp864&t=127s) The plug-in includes a dedicated text region mode designed to handle the unique high-frequency contrast patterns of typography. - [02:07](https://www.youtube.com/watch?v=xxjiQoTp864&t=127s) A real-time preview demonstrates the algorithm transforming completely illegible mobile snapshot smudges into readable text. - [02:07](https://www.youtube.com/watch?v=xxjiQoTp864&t=127s) Adobe presenters gave the standard company disclaimer that sneak peeks do not come with committed release dates or shipping guarantees. - [04:12](https://www.youtube.com/watch?v=xxjiQoTp864&t=252s) The algorithm proved effective on low-resolution, compressed internet imagery in a live restoration of a blurred photo of Adobe CTO Kevin Lynch. - [04:12](https://www.youtube.com/watch?v=xxjiQoTp864&t=252s) Lower-resolution inputs allow parameter loading, kernel calculation, and deconvolution filtering to complete almost instantaneously. - [04:12](https://www.youtube.com/watch?v=xxjiQoTp864&t=252s) Adobe MAX Sneak Peeks have a track record of maturing into shipping products, exemplified by the automated object removal demo that became Content-Aware Fill in Photoshop CS5.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=xxjiQoTp864&t=0s) Deconstructing Motion Blur: Kernels and Point Spread Functions

The demonstration opens with an Adobe researcher presenting a prototype Photoshop plug-in designed to calculate camera motion and completely reverse severe image blur. Because the tool is still in early development, the interface is crowded with intricate technical controls, prompting the presenter to load a small set of pre-configured numerical parameters rather than adjusting every setting by hand on stage. The workflow begins by pressing the “Analyze” button, triggering an intensive computational phase beneath the surface to determine exactly how the camera moved while the shutter was open.

```mermaid flowchart LR accTitle: Photoshop Deblurring Workflow accDescr: Three-stage process of the prototype plugin from blurred image to kernel analysis and final reconstruction.

subgraph Input["1. Input & Setup"] A["Load Blurred Image"] --> B["Set Region / Parameters"] end

subgraph Analysis["2. Computation"] B --> C["Analyze Heavy Lifting"] C --> D["Calculate Blur Kernel / PSF"] end

subgraph Output["3. Restoration"] D --> E["Deconvolution Filter"] E --> F["Sharp Output Image"] end ```

The direct computational output of this calculation is rendered as a small grayscale graphic called a blur kernel, or formally known in optical physics as a point spread function (PSF). As the presenter explains, this grayscale path maps the exact physical trajectory of the camera sensor during the exposure. Applying this calculated path back across the photograph reconstructs edge contrast and fine details that appeared hopelessly lost. Co-host Rainn Wilson expresses sheer disbelief as the blurred photo snaps into crisp focus, prompting a humorous exchange when he interrupts to ask how on earth the software achieved it. When told an algorithm drives the calculation, Wilson jokingly demands to know what an algorithm even is, before being told they can discuss it over Weezer later that night.

- The plug-in isolates camera shake by computing the point spread function (PSF), mapping the motion path of the camera sensor while the shutter was open. - The intensive calculation runs locally, producing a grayscale blur kernel before deconvolution filter algorithms restore the final sharp image. - “There's no way you did that,” Rainn Wilson exclaims in disbelief after watching the first blurred picture transform into a crisp photograph.

---

[02:07](https://www.youtube.com/watch?v=xxjiQoTp864&t=127s) Deblurring Unreadable Text from Mobile Photos

Moving beyond general photo enhancement, the presenter demonstrates how the technology can extract critical, unreadable data from real-world snapshots. He showcases a photograph of a poster taken on a mobile phone camera that was originally snapped to read later at home, but was so severely smeared by motion blur that every line of text became illegible. To expedite processing on the stage laptop without forcing the system to calculate the entire high-resolution frame, the presenter draws a bounding box around a specific local region of interest.

Because text characters exhibit unique high-frequency contrast patterns unlike natural landscapes, the interface requires flagging the selected zone as a dedicated text region so the algorithm can apply targeted mathematical treatments. Mid-demonstration, Wilson asks whether the feature will actually ship in the upcoming release of Photoshop, prompting the co-host to deliver the standard Adobe sneaks disclaimer: “We don't commit to timing for anything in a sneak but uh but uh you can see here.” A local preview window processes in real time, transforming the illegible smudges into crisp, perfectly readable typography. Wilson turns to the audience to demand who is in charge of Adobe's product decisions, playfully urging the team to incorporate the tool immediately while deadpanning that he personally would never pay for software because his status as a minor television celebrity entitles him to free goods.

- Cropping down to a localized bounding box allows the heavy mathematical analysis to run in seconds rather than processing the entire high-resolution image file. - The plug-in features specialized algorithmic handling tailored specifically for text extraction and edge reconstruction. - “People will really like this,” Rainn Wilson tells the product team. “People will love it. Would I pay for it? No. Absolutely not. I don't pay for anything anymore. I'm a minor television celebrity, so I get free...”

---

[04:12](https://www.youtube.com/watch?v=xxjiQoTp864&t=252s) Sharpening Kevin Lynch and Wrapping Up MAX Sneaks

For the final live test, the presenter loads an uncredited, low-quality photograph downloaded from the web showing Adobe Chief Technology Officer Kevin Lynch speaking on stage at the previous year's MAX conference. Because the image resolution is small, the parameter loading, blur kernel computation, and deconvolution run almost instantaneously. Lynch’s blurred facial features, microphone, and clothing snap into focus with dramatic clarity, drawing sustained cheers and applause from the audience.

The hosts reflect on the legacy of the MAX Sneak Peeks, reminding the crowd that past audience favorites demonstrated on the same stage—such as the automated object removal and seamless background repair that became Content-Aware Fill—ultimately shipped as flagship features in Photoshop CS5. The session concludes with the hosts thanking Rainn Wilson, collecting an autograph on stage for a colleague named Teresa, and inviting the attendees outside to the evening MAX Bash featuring a musical performance by Weezer, catering from Wolfgang Puck, and local Los Angeles food trucks.

- The final test demonstrates the plug-in's efficacy on low-resolution, compressed internet imagery using a candid shot of Adobe CTO Kevin Lynch. - Adobe previously turned sneak peek prototypes into shipping commercial software, most notably transitioning content removal research into Photoshop CS5's Content-Aware Fill. - “A couple years ago, you would have seen things like automatically removing elements from pictures and seamlessly correcting the background, which actually did ship in Photoshop CS5,” the host points out.

## Mentioned Adobe, Adobe MAX 2011, Photoshop, Photoshop CS5, Content-Aware Fill, Rainn Wilson, Kevin Lynch, Weezer, Wolfgang Puck, Peter Elst, Teresa

## Video description more Adobe MAX 2011 sneak peeks here: http://www.peterelst.com/blog/2011/10/11/adobe-max-sneak-peek-videos/

[Source](https://www.youtube.com/watch?v=xxjiQoTp864)

中文简报

Adobe展示可逆转严重相机抖动的Photoshop算法通过将曝光期间相机传感器的运动轨迹计算为点扩散函数，该原型工具能够恢复丢失的边缘对比度，从而挽救模糊照片并复原难以辨认的文本。一名Adobe研究员展示了一款Photoshop早期原型插件，它能计算快门曝光期间的相机运动，从而逆转严重的动态模糊。通过计算点扩散函数（模糊核），该工具利用反卷积滤波器恢复照片中锐利的边缘和丢失的细节。现场测试展示了该技术为相机快照去模糊、从模糊的手机照片中提取难以辨认的文本，以及锐化Adobe首席技术官Kevin Lynch的一张低分辨率网络图像。特邀联合主持Rainn Wilson登台带来诙谐解说，主持人还将该项目与此前“内容识别填充”等成功的预览技术进行了类比。 Adobe在MAX 2011大会上预览了一款实验性Photoshop去模糊原型工具，该工具通过计算传感器的物理运动轨迹，从数学上逆转严重的相机抖动。 (0:00) 该工具可计算点扩散函数（PSF）或灰度模糊核，运行反卷积滤波器以恢复边缘对比度并还原难以辨认的文字。 (0:00) 专门的设置允许对指定文本区域进行针对性的数学处理，通过局部边界框加快分析速度。 (2:07) Adobe重申其不对技术预览（sneak peeks）承诺发布计划的政策，尽管此前MAX大会上的原型（如“内容识别填充”）已成功转化为Photoshop CS5中的商用功能。 (2:07)

Original transcript

[0:00] to remove the blur and restore sharp [0:02] image. It turns out he was lucky because [0:04] we have been working on this problem for [0:06] a while and we have built a prototype to [0:09] solve it. To show it, I'll just load the [0:12] image into the Photoshop plugin we [0:15] developed. [0:17] Uh and you you can see there are a lot [0:19] of controls here. Uh because it's still [0:21] a early prototype. uh I to save some [0:24] time I just uh uh load some predefined [0:28] parameters which contain only a few [0:29] numbers. So the first thing to do with [0:31] this tool is to click the analyze button [0:34] which analyze how the image was blurred [0:36] at the first place. Uh the system then [0:39] does the heavy lifting under the hood [0:40] which involves some intensive [0:43] computation that I won't try to explain [0:45] here. Uh the output of this step is [0:49] something we call a blur kernel which [0:51] you will see in a moment. This is a kind [0:53] of slow computer. Oh now here it is. Uh [0:56] the it's essentially a grayscale image [0:59] showing how the image was blurred. [1:02] Formally it's called point spray [1:03] function but you can also view it as a [1:06] motion trajectory of the camera when the [1:08] shutter was open. Uh so next next thing [1:11] to do is to to use it to restore sharp [1:14] image. [1:17] And here is [1:22] no way. That's impossible. [1:25] That's your [1:29] way. There's no way you did that. [1:32] Yes way you [1:34] No way. No way. You [1:38] So you can see it's pretty sharp here. [1:40] Uh you can How does it do How does it do [1:42] that? [1:45] There's an al algorithm behind it. So [1:48] what the company algorithm [1:49] algorithm? Uh what's an algorithm? [1:54] We'll talk at Weezer. Okay. So yeah, if [1:58] you look at this small poster, I'm going [2:01] to switch back to original. You can see [2:03] how much improvement we have done on [2:04] this fabulous. Now, our system not only [2:08] makes blurry images look better, they [2:10] can also help recover some important [2:12] information from image. So, here is a [2:14] poster image I I shot using my cell [2:17] phone camera. Um, my thought was to [2:19] record the poster so I can read it at [2:21] home, but I only found out it's too [2:23] blurry, so I cannot read anything. So, [2:25] let's see if we can improve this image. [2:28] Again, I just load the image into the [2:30] plug-in. Uh, for now, I just load some [2:32] pre-select parameters. [2:35] Uh and because this is a relatively [2:37] large image, I'm not going to use a [2:39] whole image for analysis, which will [2:40] take some time on this computer. So I'm [2:43] just going to uh uh select a local [2:46] region [2:48] and ask the system to only analyze uh [2:51] using this local region. And this is a [2:54] spec case because the image contain [2:56] text. So we have to tell the system that [2:59] we're dealing with text region. So the [3:01] so the system knows too. I'll pass by [3:03] some special treatment. Now would this [3:06] be like in the next version of Photoshop [3:08] for real? [3:09] Remember we said sneaks [3:11] what [3:11] they may or may not appear in a future [3:13] version. Right? We don't commit to [3:15] timing for anything in a sneak but uh [3:17] but uh [3:19] you can see here [3:23] and similarly we can define another [3:25] preview window which allows you to [3:28] preview local uh results uh to see if it [3:32] is good and here it is. [3:35] You can see and [3:40] who's the boss here? Who's in charge of [3:42] all this? [3:43] Oh, it was you. You two guys, you should [3:46] do this in the next Photoshop. [3:49] People will really Seriously, I'm just a [3:51] chump. People will really like this. [3:55] People will love it. Would I pay for it? [3:59] No. Absolutely not. [4:01] I don't pay for anything anymore. I'm a [4:04] minor television celebrity, so I get [4:07] free [4:11] All right, just one more comparison [4:13] before and after. I have one last quick [4:16] demo. Uh so this is Kevin Lynch, our CTO [4:20] at last year's uh Max. Uh we we found [4:23] this image online. It's a random [4:25] internet uh photo. It's it's bad. So [4:28] let's see if we can improve it. Uh [4:32] uh because this is a relatively small [4:34] image. So I'm just going to Oh, still I [4:37] need to load some parameters. [4:39] Okay. Okay. I'm just going to analyze [4:43] this image which should be pretty fast [4:48] and restore it. So you can see it's much [4:51] better. Wow. [4:56] Thank you very much. Thank you. Wow. [5:04] pretty amazing. [5:05] So, every year you get a Photoshop thing [5:07] that just blows your mind. [5:08] Uh, we have a history of mind-blowing [5:10] Photoshop demos at Sneaks. A couple [5:12] years ago, you would have seen things [5:14] like automatically removing um elements [5:17] from pictures and seamlessly correcting [5:18] the background, which actually did ship [5:20] in Photoshop CS5. So, that's an example [5:23] of something we showed you this sneak [5:25] that's in shipping software today. Yeah. [5:27] So, yes, exactly. that it's incredible. [5:30] Well, I mean that that brings us sort of [5:32] the end of the what we have here on the [5:34] sneak. So, um before we do anything [5:36] else, please join me in thanking Rain [5:38] Wilson for being [5:46] an honor and a pleasure and I need you [5:48] to sign this for Teresa or I can't go [5:50] back to the office tomorrow. So, um, [5:53] while Rain does that, let me just remind [5:55] everyone that coming up next is the Max [5:58] Bash. You know, please go over there, [6:00] come see Weezer, our special musical [6:02] guest. We will have food, we will have [6:04] drink, food from Wolf Gang Puck, some of [6:06] the award-winning food trucks from the [6:08] Los Angeles area, all here. Don't forget [6:10] your hoodies. It's chilly out there. And [6:13] just once again, let me say thank you [6:14] for coming to Adobe Max. Thank you for [6:16] sneaks. And we will see you next year. [6:20] Good night, everybody. Enjoy weed.