Label Propagation Through Edge-preserving Filters
Supplemental Results [10.1109/ICASSP.2014.6853666]

This page contains the supplemental results for "Label Propagation Through Edge-preserving Filters". All of the images on this page are the full, high-resolution images that were provided in the paper. The results are shown in the context of depth map generation and this is presented in two ways:

  1. When overlaid on an image, a "hot" colour mapping is used that varies from black (small value) to red to white (large value).
  2. The values in the depth maps are shown using the values directly so that white corresponds to a nearby depth (large value) while black corresponds to a distant depth (small value).

Domain Transform Filter

The images below compare the outputs of the domain transform filter for two of its variants: normalized convolution and recursive form. The parameters used were $\sigma_s = 100$, $\sigma_r = 1$ and $N = 3$.

Unfiltered Image
Normalized Convolution
Recursive Form

Filtering vs Optimization

Sparse Features

Sparse Features
The depth estimates obtained by tracking sparse features through a video sequence using the method in [1].

Optimization-based Interpolation

Optimization-based Interpolation
Interpolation performed using the method in [2] to generate the final depth map.

Filter-based Interpolation

Filter-based Interpolation
Interpolation using the recursive form filter with the parameters $\sigma_s = 500$, $\sigma_r = 2$ and $N = 4$.

Iterative Filtering

A comparison between the two filter types for of $\sigma_s = 500$, $\sigma_r = \{ 0.7, 2 \}$ and $N = 4$.

Normalized Convolution
DTNC range = 0.7 DTNC range = 2
Recursive Form
DTRF range = 0.7 DTRF range = 2

User Correction

For the user-correction example, the filtering parameters were $\sigma_s = 500$, $\sigma_r = 0.7$, $N_R = 10$ and $N = 4$.

Original Depth Labelling
Original Depth Overlay Original Depth Map
Corrected Labelling
Fixed Sparse Depth Fixed Depth Map

References

  1. R. Rzeszutek; D. Androutsos, "Efficient automatic depth estimation for video," 18th International Conference on Digital Signal Processing (DSP), Santorini, Greece, July 1 - 3, 2013.
  2. R. Phan, R. Rzeszutek and D. Androutsos, "Semi-Automatic 2D to 3D Image Conversion Using Scale-Space Random Walks and a Graph Cuts Based Depth Prior", IEEE International Conference on Image Processing (ICIP) 2011, Brussels, Belgium, September 11 - 14, 2011.