Human visual system-based multi-scale tools with biomedical and security applications

Multi-scale transforms have been shown to be invaluable tools for image processing. The effectiveness of consequently formulated multi-scale algorithms have practically made them de facto standards for realizing solutions for a broad range of image processing problems. Multi-scale formulations of transforms and algorithms are motivated by the ability of the human visual system (HVS) to extract edge structures at their different scales. Image processing algorithms, consequently, have been developed which alter multi-transform coefficients of images for various means. However, the multi-scale contrasts as defined by these schemes generally not consistent with many other relevant HVS phenomena. Upon reviewing relevant HVS characteristics, new tools which are consistent with these features are presented. Accordingly, new image enhancement, image de-noising, and image fusion algorithms which make use of HVS-inspired multi-scale tools are presented as contributions to each of these fields. In this context, the aim of the presented algorithms is two-fold: The intention is to both consider new multi-scale solutions, as well as to formulate them using perceptually-driven mathematical constructs based on HVS characteristics. In the context of image enhancement, a new set of multi-scale image enhancement algorithms are presented which are able to simultaneously provide both local and global enhancements within a direct enhancement framework. For the purpose of image de-noising, a multi-scale formulation of the non-local-means de-noising algorithm is developed which is shown to both visually and quantitatively outperform existing de-noising approaches. Many algorithms to achieve image fusion based on the presented transforms are presented. One set of algorithms is based on a Parameterized Logarithmic Image Processing model, while another is based on an adaptive similarity-based weighting scheme. The interdependence between the different algorithms considered in this dissertation is also examined. A joint fusion and de-noising framework to simultaneous fuse and de-noise images is presented, as well as a patent pending system using fusion methodologies to perform various tasks. Experimental results illustrate the effectiveness of the proposed methods by both qualitative and quantitative means. The benefits of the presented methods are also validated through practical task-based evaluations in biomedical and security applications, including the automatic detection of masses in mammograms.