Infrared images combined to convert the AI to develop a color reduction is possible by AIST

National Institute of Advanced Industrial Science and technology [AIST] is 6, the infrared images to color images, AI announced. Conventional monochrome or approximate color view was the infrared night-vision image, the visible light of the color display can be, the visibility is high and.

【Here】AIST and RIKEN,AI R & D cooperative organization established in the domestic AI research and development to activation

■Increasing demand for security camera
Security awareness has increased in recent years, the security camera and surveillance camera such as security camera’s demand is increasing. At night time to shoot infrared is used, the captured image is monochrome or approximate the color is not displayed and the monitor overseer is subject to a clear picture for high visibility shooting techniques, the development of the challenge was.

Light from the light source is subject to the absorbed light and reflected light from, then depending on the wavelength, the reflectance changes color to reflect that. The visible light image and the long wavelengths of the infrared image and is,per unit area of the brightness [luminance] information is different.

Therefore, the infrared images from the natural image and converts the luminance information will not be used. Visible and infrared reflection characteristics and of the correlation is also weak for natural reproduction can be carried out by you.

■AI in the luminance information and the color information to learn
Infrared images from the natural image to convert the AIST research group focused in artificial intelligence[AI], one of the deep learning only. Language, image, or voice in a wide range of areas such as been applied to deep learning is the computer faster practical application is made possible.

The research group,the luminance information and the color information at the same time you can learn the AI model to be built. The AI model to the image feature quantity of the feature learning convolutional neural network [CNN] and videos and other time series information you can learn to cut type neural network[RNN]has been applied to.

In the conventional technology, infrared transmission or paper or cloth or other thin material to natural to convert was difficult, but this approach by the natural reproducibility is remarkably improved.

In the future, the development technique improved further, able,high-visibility security cameras, and Nocturnal Animals of ecological records, such as will be applicable and the research group is looking. [Article: Kadono 未智・The article list to look at]

Source link

Do you like this article??

Show More

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button