|
Gray-scale image with color scribbles |
Levin et al. approach |
Our approach |
|
|
|
|
|
|
|
 |
|
|
|
 |
|
|
|
|
|
|
|
 |
We compare our method with the one recently proposed which partially inspired our own: A. Levin D. Lischinski and Y. Weiss "Colorization using Optimization". ACM Transactions on Graphics, Aug 2004.
The method minimizes the difference between a pixel’s color and the weighted average color of it’s neighboring pixels. The weights are provided by the luminance channel. The minimization is an optimization problem, subject to constraints supplied by the user as chrominance scribbles. Solving it is computationally costly and much slower than our proposed technique.
The gray-scale images below are colorized both by our technique and Levin et al. approach.
|
Gray-scale image with color scribbles |
Levin et al. approach |
Our approach |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Images and scribbles originally published at: http://www.cs.huji.ac.il/~yweiss/Colorization/index.html
New examples:
 |
Original gray-scale image with user scribbles |
 |
Result using our approach CPU run-time of 1.20 sec |
 |
Result using Levin et al. approach's approach Using their supplied fast implementation of multi-grid solver CPU run-time of 22.8 sec |
 |
Result using Levin et al. approach's approach Using a slower exact Matlab least squares solver also provided by the authors. |
 |
Original gray-scale image with user scribbles. The scribbles were placed far from the edges to challenge both methods. Artificial colors where selected to better visualize the quality difference. |
 |
Our result |
 |
Result using Levin et al. approach's approach Using a slower exact Matlab least squares solver also provided by the authors.
|
 |
Result using Levin et al. approach's approach Using their supplied fast implementation of multi-grid solver
|
|
|
Original gray-scale image with user scribbles |
|
|
Result using our approach
|
|
|
Result using Levin et al. approach's approach Using their supplied fast implementation of multi-grid solver
|
Note: For fairness of the comparison all we did not change any parameter of our algorithm. For all examples the blending factor was set to be 4.
