Color Classification based on Pixel Intensity Values

What is this color? This is an easy question but sometimes difficult to answer. Because there are millions of colors but a few color names available. In this paper, we propose a method for naming colors based on pixel intensity values. Color naming is important and can be applied to many real world applications such as colorblind assistance, image segmentation, and image retrieval. Our goal here is to map color pixel intensity values to the twelve basic color names: pink, purple, red, orange, yellow, green, cyan, blue, brown, white, grey, and black. We consider only twelve categories, because basic color names were found to be shared between languages. To do so, our method firstly uses the hue value of HSV color space to categorize colors into a predefined number of ranges. Then, we apply a nearest neighbor classification on RGB color space to get color names, where only training colors containing in the same range as testing colors will be considered. Through experiments, we show that the first step will help filter out some incorrect answers, so that color names obtained from the nearest neighbor classification are more accurately. In addition, the evaluation results demonstrate that the proposed method is promising.