Noise Repair

The occurrence of noise, digital sensor noise as well as film grain noise, reduces the perceived quality of video and film significantly and lowers the viewing experience of a consumer. In times where Full HD TV (1920 x 1080) is mainstream and Ultra HD TV (3840 × 2160) is upcoming, this becomes even more evident. Furthermore, noise has an adverse influence on efficient storage utilization for long-term preservation as the high-frequency noise components require encoding with a higher bitrate, or when the bitrate is limited (e.g. in a Blu-Ray production or for satellite delivery) noise causes encoding defects (e.g. blocking).

Therefore, JOANNEUM RESEARCH is developing within the DAVID project a highly automated, highly scalable noise repair algorithm which allows archives to exploit their content properly, e.g. for TV or DVD/Blu-Ray release. Due to the high degree of automation and the good scalability, the method is expected to work for a broad range of archive content and the costs for content exploitation are expected to decrease. We plan to achieve a high degree of automation by adapting the noise repair algorithm automatically to the noise characteristics (noise magnitude, coarseness, signal dependency) occurring in the content.

An initial noise repair algorithm prototype has been developed within the first year of DAVID, it utilizes the information from the current frame (or field) and it’s temporally neighboring frames in order to reduce noise in the current frame. Novel methods have been researched which are applied within the initial noise repair algorithm in order to keep the introduction of artifacts at an absolute minimum.

First DAVID noise repair prototype. Left: original, right: noise repaired.