以艺术创作的方式挑战人脸检测算法 Creating Art while challenging face recognition algorithms

unseen-portraits

该文讲述了来自德国的两位艺术家,从如今取得巨大成功的计算机视觉项目人脸识别中获得灵感,进行艺术创作。他们使用OpenCV的人脸检测算法作为基础,让计算机在识别人脸的同时,对图像中人脸进行各种惨无人道的变形操作,以测试人脸检测的极限;而这些扭曲变形的人脸则是对应的艺术作品。视频中展现的过程更加生动有趣。

Just like other artificial intelligence disciplines, computer vision has (lots of) shortcomings. For example, despite claimed high accuracy of algorithms based on deep learning for face recognition (such as DeepFace by Facebook) there could be times wherein those algorithms could not recognize even simple images. Could you imagine this deficiency was the inspiration for an art project?

Well, two German artists namely Stephan Bogner and Philipp Schmitt have tried exploring the limitations of face recognition algorithms and produce artistic portraits as a byproduct. The project is called Unseen Portraits and it uses OpenCV library to recognize faces while producing artistic photos by manipulating the original images.

The project used two computers; one for distorting the image and one for recognition. Processing and an altered version of MapMap-Vauxhall was used for image distortion. A camera was also used to film the screen that displays distorted images and stream it to the second computer. The second computer applies OpenCV algorithm for face recognition to analyze the video in real time. As you can imagine, the algorithm initially is doing pretty well but as the image is getting distorted more and more, the capability of the software lessens. After a few moments, OpenCV algorithm could no longer recognize the photo anymore. At this very moment, the software will take a screenshot and that is the unseen portrait which is essentially invisible to the computer.

CV Dazzle has also produced a similar project. They investigated how we can conceal ourselves with the help of art in order for us to become invisible to face recognition softwares. They have teamed up with designers and stylists to trick the software. They wanted to discover how fashion could be used to camouflage from the technology of Face recognition.

The name of the CV Dazzle project was derived from the abbreviation of computer vision while Dazzle was used during WWI as a camouflage. CV Dazzle has a goal of breaking apart facial features that produces an Art that makes it unrecognizable in CV algorithms. OpenCV was also used in this project.

The difference on how humans and computers view the world was proved in these projects. The most interesting thing is that human eye (and brain) outperforms computers in recognizing distorted portraits. This fact shows that computer vision still has a lot of potential and space for improvement in the face recognition field.

Unseen Portraits from Philipp Schmitt on Vimeo.

Source from ComputerVisionOnline, posted by Amin Sarafraz.

关于dixiaocui

无人驾驶汽车研究员 半吊子摇滚混子乐手 Researcher on Intelligent Vehicle Guitarist/Singer/Rocker
此条目发表在Autonomous Driving分类目录,贴了, , 标签。将固定链接加入收藏夹。

发表评论

邮箱地址不会被公开。

此站点使用Akismet来减少垃圾评论。了解我们如何处理您的评论数据