How do image recognition algorithms work

Image recognition

  1. The Eyecam looks and moves like an eye. But why? The creator apparently wants to stimulate thought with the webcam.

    04/09/20216 CommentsVideo
  2. Azure Percept is a new edge platform with which pre-trained or your own AI models can be used as easily as possible.

  3. A rapid antigen test for Covid-19 is to be linked to an iPhone to detect the result. It is very easy to read.

  4. The software can create various motifs in different arrangements and rendering styles. Text input is required for this.

  5. In a deep fakes experiment, the ex-US president speaks to the nation. This was completely new territory for actors and MIT researchers.

    10/12/20205 CommentsVideo
  1. SARSTEDT AG & Co. KG, Nümbrecht-Rommelsdorf
  2. HCC / KPM Electronics GmbH, Siegen
  3. HEINE Optotechnik GmbH & Co. KG, Gilching
  4. operational services GmbH & Co. KG, Frankfurt am Main, Munich, Hamburg


  1. The 80 million categorized images have been created automatically. AI systems learned insults from the MIT dataset.

  2. The research team has been working on Pifu and Pifu HD for several years. The AI ​​can convert both photos and videos into 3D characters.

  3. Amazon is pausing to allow governments to formulate rules for face recognition. The company shifts the responsibility.

  4. Deep face drawing can generate faces in real time and only needs a few lines. But the AI ​​still has its limits.

    06/05/202016 CommentsVideo
  5. The database was created after the G20 summit. The deletion is not enough for the data protection officer.

Deep_fake, deep learning, neural network, AI, Prime-Air, machine learning, Tensorflow, Creative Cloud, Sundar Pichai, voice output, face recognition, Slack, software, Azure
Image recognition
  1. A group of developers created the data set for AI training and achieved good results - albeit from selfies from Instagram.

  2. Build 2020 Cognitive Services can no longer extract only English texts from images. There is also an AI to help with pronunciation.

  3. Optical character recognition is useful as long as the text is nice and straight. These systems still have problems with crooked fonts and logos. Amazon researchers want to change that by designing an intermediate step that recognizes the shape of a lettering as a frame.

  1. Willem Dafoe becomes a similar image, but one that is difficult to identify for face recognition: Researchers are developing a machine learning program that changes people's faces in such a way that they are protected as well as possible from recognition algorithms. This should also work in videos.

  2. The police's biometric database was supposed to be deleted. However, the Hamburg administrative court ruled in favor of the authority. Face recognition will continue to be used in individual cases.

  3. According to the media report, Tesla has taken over the start-up Deepscale, which specializes in image recognition. The technology could help Tesla improve autonomous driving.

  1. Microsoft has released its cloud security software Azure Sentinel. Companies can use it to monitor their network traffic. However, this costs 2.20 euros per gigabyte. A lot of traffic reduces this price drastically.

    09/25/20197 CommentsVideo
  2. In cooperation with Qualcomm, Microsoft is providing the developer kit for a small webcam that will be able to use AI models directly. Linux runs on the device.

  3. The police have more material than they can ever sift through. Artificial intelligence is supposed to help her fight against child pornography - and a breakthrough should have been achieved.
    A report by Larissa Holzki

  1. What do people look like when they are 80 years old? The Russian fun app Faceapp can show this and other effects using machine learning with a selfie. However, the software saves the source image on its own servers. In connection with the country of origin, this calls critics and analysts on the scene.

  2. Argentina is paying 30 million US dollars for a regional video surveillance system made by the Chinese company ZTE. The Chinese want to use this as a starting point for expansion in the South American market. Criticism comes from the USA.

  3. During the disruption of some Facebook services, users were able to gain insight into how the company is tagging images. The machine learning system already recognizes people, postures and emotions and can differentiate between children and adults. Nevertheless, there are problems.

  1. The success of the AI ​​nude generator Deepnude was too much for the small development team: First the download server crashed and then the team announced the end of the project. With Deepnude, with the help of GANs, images of clothed women could be converted into images of naked women.

  2. According to an Amazon patent, the company's own drones will not only deliver packages in the future, but will also be used for surveillance. Amazon customers should be able to book corresponding services.

    06/21/201921 CommentsVideo
  3. Amazon has been working on autonomous delivery drones for many years. The company has shown a new model that can already take off, land and detect objects in flight. After take-off, the drone rotates 90 degrees in the direction of flight.

    06/06/20195 CommentsVideo
  4. What does a cat shaped lab look like? Users can find out with the Nvidia tool Petswap. It is based on the Funit neural network, which is supposed to imitate human imagination - sometimes with bizarre results.

  5. GTC 2019 Using a small image editor, Nvidia introduces the neural network Gaugan. This creates rocks, water and forests from simple surfaces. At first glance, that works quite well.

    03/20/201946 CommentsVideo
  6. Although the US state of Illinois has stricter rules against the unsolicited collection of biometric data, this procedure remains legal for Google. The company creates face profiles in Google Photos to search pictures for people. This is enabled by default.

    01/02/20193 CommentsVideo
  7. In search of stalkers, the organizers used face recognition and an image database at a Taylor Swift concert. Visitors were scanned unnoticed while watching a video. According to US house rules, this is probably legal.

  8. The image recognition algorithm used by Tumblr is very imprecise: instead of nudity, it marks drawings of unicorns and the US politician Joe Biden. The company has just under two weeks until the new terms of use come into effect and the software finally decides on the content.

  9. Who needs facial recognition? A surveillance system is being developed in China that recognizes citizens by their unique gaits. The software should not be fooled by intentionally different running. But it doesn't yet work completely in real time.

  10. In order to be able to finance high-risk film projects more securely, researchers at 20th Century Fox are developing a neural network that is supposed to analyze the preferences of moviegoers. From this, the system calculates film trends for the coming year.

  11. Can software create art? A successful auction proves that there are at least buyers for it. However, the Obvious development team has probably made extensive use of other AI systems.

  12. Google rejects a lucrative deal with the US government: The company is not participating in a cloud project for the US Department of Defense. An AI cooperation with the US military was not well received by the workforce.

    09.10.201810 CommentsVideo
  13. Alibaba's announcement to enter the market for AI chips seems a little nebulous. The company is even setting up its own research team for this purpose. The Chinese also want to quickly establish themselves in the field of quantum computers.

  14. Amazon Rekognition has come under fire again: Aclu had the US congressmen analyzed by image recognition and criticized bias. The system incorrectly recognizes a disproportionate number of dark-skinned people. Amazon justifies itself.

  15. Move Mirror is a machine learning demo with Posenet that lets users move in front of the webcam and see images of other people in a similar pose - including dancers and kung fu fighters. Users can also use this to create animated GIFs.

    07/23/20181 CommentVideo
  16. 1984 and the Minority Report are now almost real scenarios. This is the opinion of Microsoft President Brad Smith. He asks questions about how facial recognition can be used sensibly and without infringing the law - and points a finger at Amazon and Google.

    07/16/20186 CommentsVideo
  17. Thanks to deep learning, Microsoft can extract the outlines of buildings from satellite images. Microsoft's map provider Bing Maps donates the data obtained in this way to the Openstreetmap community. This is not the first cooperation of this kind.

    03.07.201810 CommentsVideo
  18. Dozens of Google employees are said to have resigned in protest against the cooperation between Google and the US military. Several thousand have signed a petition against it. Now the cooperation, which expires in 2019, is not to be extended.

    06/02/201883 CommentsVideo
  19. With the help of the well-known AI software Fakeapp and Adobe After Effects, a video clip was created that lets former US President Obama speak insulting sentences. The forgery is already very successful, but not without errors. The potential of such techniques is already great.

  20. Google's AIY Vision and Voice are reminiscent of Nintendo Labo. These are craft kits made of cardboard and Raspberry Pi Zero WH. But first the Google Assistant and Tensorflow should turn the cubes into an image recognition system or a mini Google Home.

    04/17/20181 CommentVideo
  21. More than 3,100 Google employees oppose the military use of their AI. An open letter to CEO Sundar Pichai was drafted in protest against Project Maven. Long-established employees have also signed the document.

    04/09/201856 CommentsVideo
  22. With very fast motors, Playstation Eyes and open source image recognition, two developers were able to break the previous world record for solving a Rubik's Cube, also known as a Rubik's Cube in German - in 380 milliseconds. The process was recorded with a high-speed camera.

    03/08/201853 CommentsVideo
  23. Detecting enemies and positions from the air and automatically evaluating them: The US military invests a lot of money in the use of AI for military purposes. According to a report, Google provides some of the necessary technology. There should be heated discussions about it among employees.

  24. More and more users can use advanced face recognition on Facebook. The procedure is not offered in the EU or Canada - probably for data protection reasons. It could offer more security against the misuse of photos.

    02/28/20180 CommentsVideo
  25. With enough reference material, a deep learning algorithm developed by Google's subsidiary Verily can determine various characteristics of patients, such as age, gender or the risk of cardiovascular disease. The accuracy is already comparable to traditional methods.

  26. Nima should be able to assess the aesthetic and craftsmanship quality of a photo as precisely as possible without the direct reference image being available. According to Google, the results are already close to human assessment, but research is still far from productive application.

  27. Turi Create provides methods for image recognition. It should be as easy to use as possible and without prior knowledge of machine learning. Unusual for Apple: The product is open source.

  28. Rekognition Video is the new addition to AWS 'deep learning system. Faces, objects and actions can be automatically recognized in the cloud. Even celebrities should be able to be assigned in this way. The results are displayed in an overview.

    11/30/20170 CommentsVideo
  1. What can be seen in a picture? Two development teams in the USA have developed processes to automatically recognize and describe images. This is useful for image searches, but it can have other effects as well.

  2. Google searches for similar images

    Google has expanded its image search: users can now search for similar motifs in the results of the image search. Similar Images was introduced as a beta version in April.

  3. Electronic image processing
    Algorithm recognizes invisible structures

    A new image recognition algorithm has been released to the public under a free license. It makes structures invisible to the human eye visible.


  1. The Division goes into open beta

    Ubisoft has opened the open beta for The Division: Owners of an Xbox One can compete in New York from now on, PS4 and PC players from February 19, 2016. Nvidia has already released a game-ready graphics driver with optimizations for the action title.
    (The Division Beta)

  2. There is no more new smartphone for 230 euros

    The Honor 5X is another affordable smartphone with interesting features from Huawei - metal housing, fingerprint sensor and 13-megapixel camera don't sound like just 230 euros. In the test, Golem.de looked at whether the device does justice to the first good impression.
    (Honor 5x)

  3. Update for post-nuclear settlers and graphics options

    The patch to version 1.3 of Fallout 4 contains bug fixes as well as a few new functions, such as a status menu in settlements. The PC version is available now, and console players should receive the update by the end of the week.
    (Fallout 4)

  4. Nintendo presents the first application for smartphones

    What Nintendo presented as the first app for mobile games is not a real game: Players should be able to communicate with Miitomo. In addition, Nintendo Account was announced for the cross-platform exchange of, for example, saved games.
    (Miitomo)

  5. Windows 10 turns the smartphone into a desktop PC

    Microsoft brings the Continuum function of Windows 10 for smartphones: connected to an external screen, a desktop PC is created on which universal apps such as Outlook or Powerpoint run. For Continuum, however, new smartphones are required.
    (Continuum)




Did we miss something?

Email to [email protected]