The structure tensor for vector valued images is most often defined as the average of the scalar structure tensors in each band. The problem with this definition is the assumption that all bands provide the same amount of edge information giving them the same weights. As a result non-edge pixels can be reinforced and edges can be weakened resulting in a poor performance by processes that depend on the structure tensor. Iterative processes, in particular, are vulnerable to this phenomenon. In this work, a structure tensor for Hyperspectral Images (HSI) is proposed. The initial matrix field is calculated using a weighted smoothed gradient. The weights are based on the Heat Operator. This definition is motivated by the fact that in HSI, neighboring spectral bands are highly correlated, as are the bands of its gradient. To use the heat operator, the smoothed gradient is modeled as the initial heat distribution on a compact manifold M. A Tensor Anisotropic Nonlinear Diffusion (TAND) method using the spectrally weighted structure tensor is proposed to do two kind of processing: Image regularization known as Edge Enhancing Diffusion (EED) and structure enhancement known as Coherence Enhancing Diffusion (CED). Diffusion tensor and a stopping criteria were also developed in this work. Comparisons between methods show that the structure tensor with weights based on the heat operator better discriminates edges that need to be persistent during the iterative process with EED and produces more complete edges with CED. Remotely sensed and biological HSI are used in the experiments.
展开▼
