1. Sep 2010 1 Space-Time Steering Kernel Regression for Video Hiroyuki Takeda Multi-Dimensional Signal Processing Laboratory University of California, Santa Cruz Version 1

2. Sep 20102 Directory Structure Kernel Regression 3D Kernel Regression 3D The directory contains the 3-D classic and steering kernel regression for video processing. The directory contains the 3-D classic and steering kernel regression for video processing. Utilities Utilities The directory contains some utility functions. The directory contains some utility functions. Examples Examples The directory contains some video upscaling example, which show how to use the 3-D kernel regression functions. The directory contains some video upscaling example, which show how to use the 3-D kernel regression functions. MotionToolbox_Hiro MotionToolbox_Hiro The directory contains the motion estimation functions. The directory contains the motion estimation functions.

3. Sep 20103 Directory Structure Kernel Regression 3D Kernel Regression 3D The directory contains the 3-D classic and steering kernel regression for video processing. The directory contains the 3-D classic and steering kernel regression for video processing. Utilities Utilities The directory contains some utility functions. The directory contains some utility functions. Examples Examples The directory contains some video upscaling example, which show how to use the 3-D kernel regression functions. The directory contains some video upscaling example, which show how to use the 3-D kernel regression functions. MotionToolbox_Hiro MotionToolbox_Hiro The directory contains the motion estimation functions. The directory contains the motion estimation functions.

4. Sep 20104 ckr2_regular3D Description Description Second order classic kernel regression in 3-D Second order classic kernel regression in 3-D Usage Usage [z, zx1, zx2, zx3] = ckr2_regular3D(y, h, rs, rt, ksize) Returns Returns z : the estimated video z : the estimated video zx1, zx2, zx3 : the estimated gradients along x1, x2, and x3 directions zx1, zx2, zx3 : the estimated gradients along x1, x2, and x3 directions Parameters Parameters y : the input video y : the input video h : the global smoothing parameter h : the global smoothing parameter rs: the spatial upscaling factor rs: the spatial upscaling factor rt : the temporal upscaling factor rt : the temporal upscaling factor ksize : the support size of the kernel function ksize : the support size of the kernel function

5. Sep 20105 skr2_regular3D Description Description Second order steering kernel regression in 3-D Second order steering kernel regression in 3-D Usage Usage [z, zx1, zx2, zx3] = skr2_regular3D(y, h, C, rs, rt, ks, kt) Returns Returns z : the estimated video z : the estimated video zx1, zx2, zx3 : the estimated gradients along x1, x2, and x3 directions zx1, zx2, zx3 : the estimated gradients along x1, x2, and x3 directions Parameters Parameters y : the input video y : the input video h : the global smoothing parameter h : the global smoothing parameter C : the steering matrices C : the steering matrices rs: the spatial upscaling factor rs: the spatial upscaling factor rt : the temporal upscaling factor rt : the temporal upscaling factor ks : the spatial support size of the kernel function ks : the spatial support size of the kernel function kt : the temporal support size of the kernel function kt : the temporal support size of the kernel function

6. Sep 20106 ckr2_motioncomp3D_5frames Description Description Second order classic kernel regression in 3-D with motion compensation Second order classic kernel regression in 3-D with motion compensation Usage Usage [z zx1 zx2 zx3] = ckr2_motioncomp3D_5frames(y, mv, rs, h, ksize) Returns Returns z : the estimated image z : the estimated image zx1, zx2, zx3 : the estimated gradient images along x1, x2, and x3 directions zx1, zx2, zx3 : the estimated gradient images along x1, x2, and x3 directions Parameters Parameters y : the input video y : the input video mv : motion vectors mv : motion vectors h : the global smoothing parameter h : the global smoothing parameter rs: the spatial upscaling factor rs: the spatial upscaling factor ksize : the spatial support size of the kernel function ksize : the spatial support size of the kernel function Note Note This function estimate a pixel with the local analysis window, ksize x ksize x 5 frames. This function estimate a pixel with the local analysis window, ksize x ksize x 5 frames.

7. Sep 20107 skr2_motioncomp3D_5frames Description Description Second order classic kernel regression in 3-D with motion compensation Second order classic kernel regression in 3-D with motion compensation Usage Usage [z zx1 zx2 zx3] = skr2_motioncomp3D_5frames(y, mv, rs, h, C, ksize, tshift) Returns Returns z : the estimated image z : the estimated image zx1, zx2, zx3 : the estimated gradient images along x1, x2, and x3 directions zx1, zx2, zx3 : the estimated gradient images along x1, x2, and x3 directions Parameters Parameters y : the input video y : the input video mv : motion vectors mv : motion vectors h : the global smoothing parameter h : the global smoothing parameter C : the steering matrices C : the steering matrices rs: the spatial upscaling factor rs: the spatial upscaling factor ksize : the spatial support size of the kernel function ksize : the spatial support size of the kernel function tshift : the shift parameter which shifts the temporal estimating positions. (1<= tshift < 1) tshift : the shift parameter which shifts the temporal estimating positions. (1<= tshift < 1) Note Note This function estimate a pixel with the local analysis window, ksize x ksize x 5 frames. This function estimate a pixel with the local analysis window, ksize x ksize x 5 frames.

8. Sep 20108 steering3D Description Description the function estimates steering matrices for 3-D Steering kernel regression the function estimates steering matrices for 3-D Steering kernel regression Usage Usage C = steering3D(zx1, zx2, zx3, ws, wt, lambda, alpha, rs, rt) Returns Returns C : the steering matrices C : the steering matrices Parameters Parameters zx1, zx2, zx3 : the gradients along x1, x2, and x3-axes zx1, zx2, zx3 : the gradients along x1, x2, and x3-axes ws : the spatial window size of the estimation of steering matrices ws : the spatial window size of the estimation of steering matrices wt : the temporal window size of the estimation of steering matrices wt : the temporal window size of the estimation of steering matrices lambda : the regularization parameter lambda : the regularization parameter alpha : the structure sensitivitity alpha : the structure sensitivitity rs : the spatial downsampling factor rs : the spatial downsampling factor rt : the temporal downsampling factor rt : the temporal downsampling factor

9. Sep 20109 Directory Structure Kernel Regression 3D Kernel Regression 3D The directory contains the 3-D classic and steering kernel regression for video processing. The directory contains the 3-D classic and steering kernel regression for video processing. Utilities Utilities The directory contains some utility functions. The directory contains some utility functions. Examples Examples The directory contains some video upscaling example, which show how to use the 3-D kernel regression functions. The directory contains some video upscaling example, which show how to use the 3-D kernel regression functions. MotionToolbox_Hiro MotionToolbox_Hiro The directory contains the motion estimation functions. The directory contains the motion estimation functions.

10. Sep 201010 Examples 8 examples are available to show how to use the kernel regression functions. 8 examples are available to show how to use the kernel regression functions. “Stefan_3DSKR_motioncomp.m” “Stefan_3DSKR_motioncomp.m” Space-time upscaling example of Foreman video using the 3- D steering kernel regression with motion compensation. Space-time upscaling example of Foreman video using the 3- D steering kernel regression with motion compensation. “Foreman_3DSKR_motioncomp.m” “Foreman_3DSKR_motioncomp.m” Space-time upscaling example of Foreman video using the 3- D steering kernel regression with motion compensation. Space-time upscaling example of Foreman video using the 3- D steering kernel regression with motion compensation.