- Conference date: 21–26 October 2007
- Location: Campinas, São Paulo (Brazil)
The objective of the paper is to determine the coordinates of the dominant colors in complex digital images. The key issue is the selection of an appropriate color space and optimal quantization of the selected color space. Five color spaces, RGB, LCh, Lab, I1I2I3 and HSI, and several quantization levels are involved in evaluating the performance of the propose method. The method is based on the local detection of the maxima of the color histogram. It is analyzed how the quantization and the size used in the localization of the maxima affect to the number of dominant colors obtained and their color coordinates. First, a group of twenty eight images has designed itself test. The objective of using images test is to be able to compare the coordinates of the dominant colors that the method provides with the expected values. This allows evaluating what color space provides smaller mean square error in the calculation of the color coordinates of the dominant colors, as well as their area. They have been designed varying the luminance, the hue and the chromes covering the approximate range of the real scenes to process. The and I1I2I3 color space was found to perform better than the other color spaces tested, independently of the hue and chromes of the present objects in the scene. Finally, results obtained with real images are shown.
Data & Media loading...
Article metrics loading...