MICCAI ́06 Workshop on Biophotonics Imaging for Diagnostics and Treatment

In this theoretical paper we address the question of how to encode the local signal variation of multidimensional, multispectral signals. To this end, we first extend the concept of intrinsic dimension to the case of multispectral images in a way which is not depending upon the chosen colour space. We then show how additive, multiplicative, and occluded superpositions of oriented layers can be detected and estimated in multispectral images. We expect our results to be useful in applications that involve the processing of multispectral images, e.g. for feature extraction, compression, and denoising. Moreover, our methods show how the detection and estimation of features like orientations, corners, crossing etc. can be improved by the use of multispectral images.