Wrapper Approach to Select a Subset of Color Components for Image Segmentation with Photometric Variations

The choice of a color model is of great importance for many computer vision algorithms. However, there are many color models available; the inherent difficulty is how to automatically select a single color model or, alternatively, a subset of features from several color models producing the best result for a particular task. To achieve proper colors components selection, in this paper, it was proposed the use of wrapper method, a data mining approach, to obtain repeatability and distinctiveness in segmentation process. The result was compared with neural network method and yields good feature discrimination. The method was verified experimentally with 108 images from Amsterdam library of objects images (ALOI) and 10 aerial images with different photometric conditions. Furthermore, it has shown that the color model selection scheme provides a proper balance between color invariance (repeatability) and discriminative power (distinctiveness).