2011 年 125 巻 p. 73-82
For the purpose of developing the computer based automated navigational watch system, this paper tried one of the basic techniques of computer vision, color based pixel classifier, to determine whether a pixel is sea surface or not. The technique is based on Bayes rule and use RGB color histogram calculated by labeled photos. Using 128 photos for learning, our experiment showed that the Bayes classifier detects sea pixels with 80% true positive and 25% false positive. The result is sufficient to use the classifier as the one constituting the vision algorithm of our watch system.