Optical Flow Diffusion with Robustified Kernels

Structural flow smoothing for shape interpolation

“…The block matching algorithm (BMA) [3], associates a block of pixels from a reference image I t+1 with blocks of pixels from the initial image I t , and chooses maximum correlation correspondence :…”

“…The Hessian recording the second order derivatives represents a measure of variation in the local data geometry. The Hessian can be used as a detector of change in the direction of the optical flow [3], and can be also applied to the DBMA flows. The Hessian of a vector field is represented as :…”

“…The Hessian indicates major changes in the direction of the structural flow, detecting the boundaries of various artifacts in the volumetric image. Hessians have been used as kernels for diffusion based smoothing in various applications [3,9]. In this study we employ the Hessian of the local data as the covariance matrix in the Heat kernel from (4).…”

“…Diffusion algorithms do not eliminate the influence of outliers, but rather diffuses them in the neighbourhood, causing a bias in the estimation of the structural flow. A solution to the problem of diffused outliers is to combine robust statistics with the diffusion kernel [3,7]. Order statistics algorithms have the ability to eliminate up to 50 % outliers and have been successfully used for image filtering.…”

“…The structural flow is similar to the optical flow estimated from image sequences [3]. The vectorial field in this case is used to model correspondences between consecutive slices.…”