Blind Separation ofImageSignals withNoiseDetection andEstimation
Xiaowei Zhang, Jianming LuandTakashi Yahagi · 2006
We propose anindependent component analysis whenanormal pixel appears, theupdating methodwillbe (ICA) approach whichisrobust against impulse noise. Itconsists switched tothefixed-point algorithm (2). We will thenobtain ofnoise detection andimagesignal separation. We introduce a theseparation matrix torecover source imagesignals. This self-organizing map(SOM)network todetermine iftheobserved imagepixels arecorrupted bynoise. We markeachpixel to paperutilizes theadvantages ofboth(2)and(3)butonly distinguish normalandcorrupted ones. After that, weuseone updates a single separation matrix. Thesimulation results oftwotraditional ICAalgorithms (fixed-point algorithm and willshowthat, theproposed approach outperforms thetwo Gaussian moments-based fixed-point algorithm) toseparate the original proposed algorithms withimpulse noise interference images. Theproposed approach hasthecapacity torecoverinimages themixedimages andreduce noise fromobserved images. The simulation results showthattheproposed approach issuitable forpractical unsupervised separation problem.