Archetypal coherent color acquisition: An experimentation in the field of color based object recognition

We propose an efficient approach of invariant color selection from an object possessing various transitions in color. We experiment on database of 200 guava samples possessing three classes; Raw, Ripe and Overripe. Representative color is chosen based on the concept that the dominant color of the object is representation of a dominant hue in the image. CIE L-a-b domain has been used to serve the purpose. The most consistent hue of the image along with its corresponding lightness and saturation value qualifies as the dominant color. 92.5% accuracy has been achieved by training radial centers over 80% of the dataset. We conduct comparative study of feature selection based on color moments and RGB histogram vector, both lagging our method by yielding 80% classification accuracy.