Question

I’m implementing an automatic color cast removal based on [1], which seem like a robust, simple and yet well-performing method. To avoid removing an intrinsic cast by a predominant color such as large regions of vegetation, or water, they use a method of image annotation described by [2].

The color cast detector use a multiclass support vector machine to classify image regions as sky, skin, vegetation, water or other. My problem is that [2] only describes the method, they do not include the parameters of the hyperplanes resulting from training the SVM. Training a new SVM is way out of my scope, but I haven’t found any similar works including ready-to-use data. I would really appreciate one of the following:

A. A set of hyperplane parameters resulting from training using the method in [2].

B. Some other image annotation method for sky/skin/vegetation/water, including trained parameters or not requiring training.

C. Some free image database containing annotated regions of sky/skin/vegetation/water, that I can use to train a new SVM using the method in [2].

References

1. F. Gasparini and R. Schettini “Color Balancing of Digital Photos Using Simple Image Statistics“
2. C. Cusano, G. Ciocca and R. Schettini “Image annotation using SVM“

Answer 1

I got an answer from one of the authors, they don’t have any code or data left. Instead he pointed me to colorconstancy.com, which contains links to both source code and different image databases. He also mentioned web.mit.edu/torralba/www/, containing an annotated image database that I will use for training the algorithm.