From detecting to recognizing color codes

In this paper we present efficient ways of detecting color codes in field images and recognizing the colors to read their codes. The color-code tag used in this paper consists of the code area which contains concentric, multiple circular rings on the white background. To generate the test images, we used two cameras of different brands, that is, a Canon and a Nikon. The images were taken outdoor in daylight, and the resolution was 2544 times 1696 (= 4 Mpixels). The images of figures were taken as waist shots. In the detection, we used the relative characteristics of white color in RGB color space and in the saturation domain of HSI color space. In the recognition, we used k-means clustering algorithm in the a*b* plane of L*a*b* color space to segment rings of different colors, and used hue values to determine their colors. The total number of images used was 144. The detection rate was around 99.3% and the recognition accuracy for the detected codes was around 97.8%