A combined alignment and registration scheme of lesions with psoriasis

A scheme that registers and aligns digital image lesions of psoriasis within and between sessions is proposed. Lesions to be tracked are found under the assumption of being the object of largest size in thematic maps produced by a two-step hierarchical classification scheme that uses the output of an expectation-maximization algorithm to obtain a classification window of optimal size. Advantage is taken of the fact that the shape and the size of lesions with psoriasis do not change very much along the time. A first alignment of the lesions is done assuming that the correspondence between points is given by equivalence of positions of pixels after translations and rotations. Finally, a combined contextual registration and alignment scheme is applied. The alignment and registration schemes both use an Extreme Value Detection Algorithm based on a retinal mapping model. The output of the scheme is satisfactory not only in terms of visual appreciation of the aligned lesions, but also because the variation within sessions in the aligned lesions resembles randomly distributed Gaussian noise.