An adaptive image segmentation algorithm for natural scene images based on probabilistic neural networks

Considering the problem of automatic objects detection in natural scene images, a new adaptive image segmentation algorithm for natural scene images based on probabilistic neural networks (PNN) is proposed in this paper. The probabilistic neural networks used for image segmentation is designed in this algorithm. Its input vector consists of image global color features computed by the statistics theory and pixel´s gray-values in R, G and B color channels. Its output vector is the binary vector including four elements, and the element with value 1 represents recognized color category. This neural network can divide all pixels of one color image into four types as yellow, red, blue and others. Then the color image will be segmented to the binary image corresponding to given color. To verify the validity of proposed algorithm, it is applied to many natural scene images. Experimental results show the algorithm can efficiently segment object regions of specific colors from color images, and is more adaptable and reliable to natural lighting conditions. The proposed algorithm has important application value in object detection and recognition from natural scene images.