In a previous work, we developed a distance-based formalism adapted to image processing. In colour, this formalism is correlated to human perception using an adapted distance function. In this work, we extend this framework to multispectral domain. The goal is to construct an image processing formalism guided by a physical point of view. In this context, a suitable distance function between two spectra has to be specified and selected. In this paper, we specify essential characteristics of multivalued distance using two linear transformations. Different multivalued distance behaviours are compared using Gaussian distribution. Next we propose a modified distance using linear transforms. Then we compare the behaviours of different distances on real spectral images after morphological process.