Many approaches of texture analysis by color or hyperspectral imaging are based on the assumption that the image of a texture can be viewed as a multi-component image, where spatial interactions within and between components are jointly considered (opponent component approach) or not (marginal approach). When color images are coded in multiple color spaces, texture descriptors are based on Multi Color Channel (MCC) representations. By extension, a Multi Spectral Band (MSB) representation can be used to characterize the texture of material surfaces in hyperspectral images. MSB and MCC representations are compared in this paper for texture classification issues. The contribution of each representation is investigated with marginal and/or opponent component strategies. For this purpose, several relevant texture descriptors are considered. Since MSB and MCC representations generate high-dimensional feature spaces, a dimensionality reduction is applied to avoid the curse of dimensionality. Experimental results carried out on three hyperspectral texture databases (HyTexiLa, SpecTex and an original dataset extracted from the Timbers database) show that considering between component interactions in addition to the within ones significantly improves the classification accuracies. The proposed approaches allow also to outperform state of the art hand-designed descriptors and color texture descriptors based on deep learning networks. This study highlights the contribution of hyperspectral imaging compared to color imaging for texture classification purposes but also the advantages of color imaging depending on the considered texture representation