Physically Consistent Variational Denoising of Image Fluid Flow Estimates

Abstract. High-speed image measurements of fluid flows define an important field of research in experimental fluid mechanics and the related industry. Numerous competing methods have been developed for both 2D and 3D measurements. Estimates of fluid flow velocity fields are often corrupted, however, due to various deficiencies of the imaging process, making the physical interpretation of the measurements questionable. We present an algorithm that accepts vector field estimates from any method and returns a physically plausible denoised version of it. Our approach enforces the physical structure and does not rely on particular noise models. Accordingly, the algorithm performs well for different types of noise and estimation errors. The computational steps are sufficiently simple to scale up to large 3D problems in the near future.

• Type: Conference
• Year: 2008
• Where: DAGM
• Authors: Andrey Vlasenko, Christoph Schnörr
• Keywords: DAGM 2008, Fluid Flow, Fluid Flow Velocity, Image Processing, Numerous Competing Methods