General Adaptive Neighborhood Image Processing for biomedical applications

In biomedical imaging, the image processing techniques using spatially invariant transformations, with fixed operational windows, give efficient and compact computing structures, with the conventional separation between data and operations. Nevertheless, these operators have several strong drawbacks, such as removing significant details, changing some meaningful parts of large objects, and creating artificial patterns. This kind of approaches is generally not sufficiently relevant for helping the biomedical professionals to perform accurate diagnosis and therapy by using image processing techniques. Alternative approaches addressing context-dependent processing have been proposed with the introduction of spatially-adaptive operators (Bouannaya and Schonfeld, 2008; Ciuc et al., 2000; Gordon and Rangayyan, 1984;Maragos and Vachier, 2009; Roerdink, 2009; Salembier, 1992), where the adaptive concept results from the spatial adjustment of the sliding operational window. A spatially-adaptive image processing approach implies that operators will no longer be spatially invariant, but must vary over the whole image with adaptive windows, taking locally into account the image context by involving the geometrical, morphological or radiometric aspects. Nevertheless, most of the adaptive approaches require a priori or extrinsic informations on the image for efficient processing and analysis. An original approach, called General Adaptive Neighborhood Image Processing (GANIP), has been introduced and applied in the past few years by Debayle & Pinoli (2006a;b); Pinoli and Debayle (2007). This approach allows the building of multiscale and spatially adaptive image processing transforms using context-dependent intrinsic operational windows. With the help of a specified analyzing criterion (such as luminance, contrast, ...) and of the General Linear Image Processing (GLIP) (Oppenheim, 1967; Pinoli, 1997a), such transforms perform a more significant spatial and radiometric analysis. Indeed, they take intrinsically into account the local radiometric, morphological or geometrical characteristics of an image, and are consistent with the physical (transmitted or reflected light or electromagnetic radiation) and/or physiological (human visual perception) settings underlying the image formation processes. The proposed GAN-based transforms are very useful and outperforms several classical or modern techniques (Gonzalez and Woods, 2008) - such as linear spatial transforms, frequency noise filtering, anisotropic diffusion, thresholding, region-based transforms - used for image filtering and segmentation (Debayle and Pinoli, 2006b; 2009a; Pinoli and Debayle, 2007). This book chapter aims to first expose the fundamentals of the GANIP approach (Section 2) by introducing the GLIP frameworks, the General Adaptive Neighborhood (GAN) sets and two kinds of GAN-based image transforms: the GAN morphological filters and the GAN Choquet filters. Thereafter in Section 3, several GANIP processes are illustrated in the fields of image restoration, image enhancement and image segmentation on practical biomedical application examples. Finally, Section 4 gives some conclusions and prospects of the proposed GANIP approach