Modified convex divergence ICA for separation of mixed images
Durai Sugumar, Ponnusamy Thangapandian Vanathi, Alexandra Jasmine · 2014
The acquired images in real time scenario are mixed in most cases. The separation of these mixed images is very critical since the number of components mixed and the pattern of mixing are unknown. With these unknown parameters, BSS (Blind Source Separation) plays a vital role in separation of the image mixtures. ICA, one of the widely used techniques of BSS provides better separation by finding the independency between the sources to separate the mixtures. It uses different contrast function for finding the independency. The contrast function is based on the convex divergence measure in Convex Divergence ICA (CDIV-ICA). The aim of this paper is to achieve faster and accurate separation which is accomplished by the modified CDIV-ICA. The proposed method uses a modified contrast function to find the independency between the different components. The parameters like SIR and the execution time are analysed for various image mixtures in MATLAB. From the simulated results, the modified algorithm produces better SIR compared to the Convex Divergence ICA algorithm. The algorithm converges faster in finding the demixing vector to separate the components compared to other methods. 28 image mixtures of generated database and 2 real time dual energy chest X ray image mixture are used for the experiment and the results are discussed and presented.