Removing noise from images is fundamental not only for improving the quality of the digital sensors, but also as a support for many knowledge extraction and image analysis approaches that are traditionally affected by non-Gaussian noise. In digital acquisition, noise and demosaicking are strictly correlated by a cause-effect relationship.
This project will develop fast denoising and demosaicking algorithms necessary to establish a universal image processing chain, applicable to all raw digital images obtained from reflex or compact off-the-shelf cameras, and to more specialized image generation systems, including digital photon capture array (CCD or CMOS). Denoising and demosaicking are tightly related topics and it is expected that performing both steps in a single process (instead of independently performing denoising after demosaicking) will improve the overall results. This idea is at the basis of this project, where we exploit the peculiarities of the demosaicking algorithm for improving denoising.
The project’s team already showed a high synergy of capacities and backgrounds during a decade of profitable scientific collaboration on image analysis targeting embedded systems with particular emphasis on videos and image deinterlacing.
The team as a whole already produced a number of journal articles in high quality International Journals and strengthened the collaboration co-organizing workshops, conferences, seminars and visiting the respective Institutions regularly.
As separated entities, the Italian team has an historical Industrial collaboration with STMicroelectronics for the design of high quality video sensors capable of executing advanced image enhancement algorithms while the Korean members of the team has a very strong background on image processing for embedded systems and a long time collaboration with Samsung. The achievement of the project will be disseminated to the scientific communities as well as to the industrial partners.
The project is structured into four workpackages and three crucial milestones for the three years.
WP1 Demosaicking (Korean unit) [M1-M18]: State of the art and development of demosaicking, and acquisition models.
WP2 Denoising models and algorithms (Italian unit) [M1-M18]: State of the art and development of denoising and noise models.
WP3 Combining demosaicking and denoising (both) [M10-M34]: Drive the achievements of WP1 and WP2 through the goal of mixing demosaicking and denoising.
WP4 Dissemination (both) [M6-M36]: Dissemination the project results, organization of events. Creation of web project’s demonstration where users can upload their own images to test the algorithms.
M1 [M6] Technical analysis of existing studies. Algorithm to obtain noiseless CCD image. Noise model development for demosaicking. Correlation study between the noisy image and noiseless image.
M2 [M24] CFA pattern development and performance analysis. Analysis of existing demosaicking and denoising methods. Study combination of demosaicking and denoising. Removing noise during super resolution image generation process. White balancing and gamma correction effects on noise.
M3 [M36] Noise suppress method during color format conversion. Performance analysis and comparisons. Design of CFA and its demosaicking method development. Dissemination of the final results.
Meeting 1: JDEM (Joint demosaicking and denoising in digital images)
June 26, 2016, Sunday – Incheon University (Korea)
Meeting 2: JDEM (Joint demosaicking and denoising in digital images)
July 11, 2016, Monday – Computer Science Department – Università degli Studi di Milano (Italy)
Meeting 3: JDEM (Joint demosaicking and denoising in digital images)
July 13, 2017, Thursday – Computer Science Department – Università degli Studi di Milano (Italy)