Color Cast Detection Method Based on Multi-feature Extraction - Advances in Image and Graphics Technologies - page 108

as follows: 1. Histogram of RGB color space. 2. Statistical characteristics of the Ga-
bor filter. 3. Combination of histogram of RGB color space and statistical characteris-
tics of the Gabor filter. 4. All features combined
Tab. 3 is the result of the values of the statistics methods under 460 iterations.

Table 3. Classification results of different feature extraction methods with Adaboost

R R R

No.

1 0.9707 0.1257 0.0520

2 0.9320 0.3338 0.1306

3 0.9753 0.1198 0.0471

4 0.9735 0.1100 0.0462

From Tab. 3, we can see that all values of proposed methods are higher than 93%
and the combination of all features in the paper is the best. We choose the first twome-
thods in Tab. 3 to do the iteration test. The tendency chart is shown in Fig. 3.

Fig. 3. Error rate curve with different iterations

As we can see in Fig. 3, with the increasing number of iterations, the error rate is de-
creasing distinctly. What's more, the fourth method is a little better than the third one.

5 Conclusion

We developed a new color cast detection method which is based on the multi-feature
extraction and shown that the method is universal to natural images.The experiments
in the paper have shown that the method is pretty well with high detection rate and
low error rate. So we can conclude that the method is suitable for the color cast detec-
tion before color correction and can help reduce the situation of over-correction. But
because of the quantitative limitation of the standard color images, the experiment
result is not as good as we think, in the next step, we can extend the numbers of the
samples to get a better result.