On September 16, in the Internet age of increasingly sensitive personal information, Internet users tend to use "fuzzy" or "mosaic" to mask sensitive information in the picture. However, this may no longer be applicable after, researchers at the University of Texas has developed a set of "anti-mosaic" means to break mosaic and fuzzy block, correctly identified the picture information.

People photo mosaic of smearing is not complete, human eyes do not recognize the information below does not mean that machines do not recognize. Researchers created a software for facial recognition and text recognition system can accurately identify the content is covered by blur or mosaic. Ability to recognize and it quite tough, whether it's fuzzy video on Youtube, or mosaic figure, is self-evident. But to note here is that this set of tools is currently only "recognize" rather than "restitution". Which means it can match in a bunch of images and mosaics below matches the contents of the image, but not really "anti-mosaic".

Despite such regret, but the launch of this technology is quite shocking. Because it can be so easy to be realized. Developers use the Torch, which is a open source software that can easily available on the Internet. Used to test images from the network. Debugging software researchers to identify fuzzy pictures of before and after, and easily reached a high degree of accuracy.

Experiment, the recognition rate of the computer fuzzy video on Youtube 80~90%. Even with the most serious painting mosaic images, accuracy can reach 50~70%. Pictures that have been modified by means of P3 identify the lowest rate, with only 17%. P3 (Privacy-Preserving Photo Sharing) was developed at the University of Southern California an image encryption tool, designed for live picture spread in the Internet age. The pictures in JPEG format has a very good encryption.

Although you do not have to worry for of your own are spread on the Internet, but the advance guard is not totally unfounded. Experimenter earlier just using off-the-shelf software, and did it without much effort on 90% recognition accuracy is achieved. And because the software is open source, it also means that any thinking person, maybe a little effort can achieve high recognition accuracy. (Although complete recovery is not real) is worried about disclosure of personal information of Internet users, the researchers suggested was: deprecated "fuzzy" and "mosaic" in favour of completely blacked out to protect their privacy. Although a black square on the photo is not very beautiful, but to protect the safe and quite effective.

9月16日消息，在个人信息日益敏感化的网络时代，互联网用户一般都会习惯用“模糊”或“马赛克”来遮盖照片上的敏感信息。不过这一手段以后可能不再适用了，得克萨斯大学的研究者日前开发出了一套“反马赛克”手段，可以突破马赛克和模糊的遮挡，准确地辨认图片信息。

一般人对照片进行的马赛克涂抹并不彻底，人眼辨认不出其下的信息不代表机器也辨认不出。研究者建立了一个用于面部识别和文字识别的软件系统，可以相当准确地识别被模糊或马赛克遮掉的内容。而且它识别能力相当强悍，无论是Youtube上的视频模糊，还是马赛克修图，都不在话下。不过这里要说明的是这套工具目前还只能“识别”而不能“复原”。也就是说它能在一大堆图像中匹配到和马赛克下面的内容相符的图像，而无法做到真正的“反马赛克”。

尽管有如此的遗憾，不过这项技术的面世还是令人相当震惊。因为它可以如此简单地被实现。开发人员使用了Torch，这是一个可以在网上轻易获得的开源软件。用来测试的图片也都来源于网络。研究人员调试软件对模糊前后的图片进行辨认，结果轻易达到了很高的准确度。

实验中，计算机对Youtube模糊视频的识别率达到了80～90%。而即使经过最严重涂抹的马赛克图片，识别准确度仍然能达到50～70%。而通过P3手段修改过的图片识别率最低，只有17%。P3（Privacy-Preserving Photo Sharing）是南加州大学开发的一个图片加密工具，专为互联网时代的图片传播而生。对JPEG格式的图片有很好的加密效果。

虽然目前大家还不必为自己散布在网上的图片感到担心，不过提前做好防范并非杞人忧天。前文中的实验人员仅仅使用现成的软件，没费多大力气就实现了90%的识别准确度。而且由于软件是开源的，也就意味着任何有想法的人，也许稍加努力就可以达到更高的识别准确度。（虽然彻底复原并不现实）对于十分担心自己个人信息泄露的网民，研究人员的建议是：弃用“模糊”和“马赛克”，改用彻底涂黑的方法来保护自己的隐私。虽然照片上一个黑方块不太美观，不过用来保护安全相当有效。