2018/5/29: add
-
more efficiency process separation
-
Warping Process
-
Detph Fill
-
Sprite Update -[x] Remember the hole location
-
Image Update
-
Initial Fill
-
Texture Refine
- BG(BackGround) is static
- For the moving BG, motion estimation is required to compensate the BG motion, then can use this function to filling the hole.
a video with associated depth map
- step 1 : Warping process
- step 2 : Depth Fill
- step 3 : Sprite Update
- step 4 : Image Update
- step 5 : Initial Fill
- step 6 : Texture Refinement
This is the algorithm to generate the a 3D image warping image and it's associated depth map, there have distortion area in both Warpied image and it's associated depth map
see more detail in package Warping
Depth Fill is the pre-process of the depth map, which can fill the distortion area of the depth map with K-means Algorithm.
It's purpose is to separate the FG(foreground) and the BG(background).
Choose i's neighborhood which is like a M * N squared area. using K-means(K=2) to cluster those pixel's into to region. then we can get the cluster center named c_min
and c_max
Then just do a judge function to change the i's value. If it is big than C_min
, it will be C_min
. If it is less than C_min
, it will not change. See more detail in package 'DepthFilling' and the paper.
Sprite update
contain the BG sprite updation and DM sprite updation.
BG and DM sprite are only one in whole process. just to calculate the median C_min
and the do judge function to distribute the pixel which is needed to store in the Sprite
Need to be done
initial Filling contains two method, filling small holes just to using Laplace equation to fill. Fill large hole is using author's previous work( 😭 I don't reproduce this work)
Texture refinement need to be done