Neural network system for texture analysis (S. Lu)

Many images to be processed are textured images such as medical images, landscape images, and mineral images. Textured images are classified and segmented using 2-D autoregress (AR) model and neural network. The network consists of three subnets: input subnet, analysis subnet, and classification subnet. The network is used to both establish a 2-D AR model for a texture region and to implement region identification and segmentation. Good performance is achieved by an adaptive learning process which accurately estimates an AR model for a texture. After the learning, a frame from textured image is input and forwarded directly through the neural network, the computation time for texture feature extraction is dramatically reduced.

Revised: 94.10.25