Automatic Multilevel Color Image Thresholding by the Growing Time Adaptive Self Organizing Map

This paper presents a simple but effective algorithm for color image quantization. A growing time adaptive self organizing map (GTASOM) is used to find the color distribution in the three-dimensional RGB color space. The mapping property of self organizing maps from a high dimensional space to a lower dimensional grid is the basic idea in this work. A GTASOM network is trained with the pixel's colors of the image to find the input distribution. The number of neurons is increases in this network with some criteria to have a good representation of input data distribution. Then a peak finding algorithm is applied on the network weights to find the quantization levels automatically. The experimental results show the good superiority of the proposed algorithm