[ Camouflaging Objects ]
Tanachat Nilanon (tnilanon)
This project makes use of the paper "Color Transfer between Images". Also here is the slides for the presentation given in class.
The task can be splitted into two parts.
Assumption: Similar things can be replaced
Use mask to limit similarity checking to just the object of interest
Assumption: "Sharp" changes signify an object
Don't want sharp changes
Only consider descriptors near the border of the mask
More complicated approach, but has more potential, since we can then handle automatic rotation and scaling
A cat and its mask
 |
 |
Cross-correlation
| Original | Copy over | Color transfer |
|---|---|---|
 |
 |
 |
 |
 |
 |
| Different location + Color transfer |
|---|
 |
A snake and its mask
 |
 |
Cross-correlation
| Original | Copy over | Color transfer | Color transfer (Whole image as base) |
|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
A few planned implementations are still not complete. This includes the Feature Descriptor approach for finding location for the object
The color transfer method blends the object into the background too much. We could do a weighted average between the original object and the color-tranferred object, but the weights are still needed to be determined. Gradient domain blending could also work.
I would like to thank Prof. Efros for teaching and making this course awesome. This course will be totally different without his encouragement and sense of creativity. In addition, I also would like to thank our TA, Natasha, for being understanding and helpful. Last but not least, it cannot be deny that fellow students are an integral part to class experience. You guys are wonderful.