Neural networks for the texture classification of temporally consistent segmented regions of FLIR sequences
John F. Haddon, James F Boyce · 2002
Texture can be interpreted as a measure of the edginess about a pixel and can be described by edge co-occurrence matrices. When the matrix is decomposed using discrete 2D orthogonal Hermite functions, the coefficients are a low order feature vector which is characteristic of the texture. They can be used as features in a neural network classifier for labelling regions of FLIR images segmented using co-occurrence based techniques. Emphasis is placed on ensuring temporal consistency of the segmentation.