Texture classification of segmented regions of FLIR images using neural networks

The major regions in a sequence of FLIR images, segmented using co-occurrence techniques, are subjected to texture classification. Texture can be thought of as a measure of the edginess about a pixel and can thus be described by edge co-occurrence matrices. These are temporally smoothed to ensure consistency in segmentation between the images of a sequence. The matrix of each segmented region is decomposed using 2-dimensional orthogonal Hermite functions. The coefficients of the decomposition are regarded as a low order feature vector which is characteristic of the region texture. Principal component analysis is used to determine the optimum combination of coefficients. In an application for use with infrared imagery these are then used to train a range of multilayer perceptron neural networks using autonomously segmented regions in every fifth image. The best network is then used to classify the remainder of the 300 images