eduzhai > Applied Sciences > Engineering >

CIE XYZ Net Unprocessing Images for Low-Level Computer Vision Tasks

  • Save

... pages left unread,continue reading

Document pages: 15 pages

Abstract: Cameras currently allow access to two image states: (i) a minimally processedlinear raw-RGB image state (i.e., raw sensor data) or (ii) a highly-processednonlinear image state (e.g., sRGB). There are many computer vision tasks thatwork best with a linear image state, such as image deblurring and imagedehazing. Unfortunately, the vast majority of images are saved in the nonlinearimage state. Because of this, a number of methods have been proposed to "unprocess " nonlinear images back to a raw-RGB state. However, existingunprocessing methods have a drawback because raw-RGB images aresensor-specific. As a result, it is necessary to know which camera produced thesRGB output and use a method or network tailored for that sensor to properlyunprocess it. This paper addresses this limitation by exploiting another cameraimage state that is not available as an output, but it is available inside thecamera pipeline. In particular, cameras apply a colorimetric conversion step toconvert the raw-RGB image to a device-independent space based on the CIE XYZcolor space before they apply the nonlinear photo-finishing. Leveraging thiscanonical image state, we propose a deep learning framework, CIE XYZ Net, thatcan unprocess a nonlinear image back to the canonical CIE XYZ image. This imagecan then be processed by any low-level computer vision operator and re-renderedback to the nonlinear image. We demonstrate the usefulness of the CIE XYZ Neton several low-level vision tasks and show significant gains that can beobtained by this processing framework. Code and dataset are publicly availableat this https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×