MULTISCALEINTENSITYESTIMATIONFOR MARKED POISSONPROCESSES

Rebecca Willett

Abstract

Inference onseverely data-starved Poisson processes canbedramatically improved byusing auxiliary information about measured discrete events intheformofmarks. Marksarewidely available inmanyapplications, andcantake theformofphoton energy, time delay information, packet size, orother forms ofcharacterizations. Effectively using marks results ininnovative signal processing methodsanddramatic error reductions forPoisson intensity estimation. Theefficacy oftheproposed method isdemonstrated inthecontext ofphoton-limited spatio-spectral intensity estimation. IndexTerms-Poisson processes, Wavelets transforms, Multidimensional signal processing 1.MARKED POISSONPROCESSES MarkedPoisson processes arise whenindirect observations ofacomplex physical phenomenon arecollected bycounting discrete events andmeasuring anassociated auxiliary feature vector foreach event. Suchinformation isavailable influorescence microscopy, hyperspectral imaging, solar spectroscopic imaging, andother applications. Multiphoton laser scanning microscopy (MPLSM), forexample, isacutting-edge tool for high-resolution imaging ofliving tissue andorganisms used tostudy cellular dynamics [1].Theresult ofthis powerful technique isaseries ofdetected photons resulting fromfluorescence events occurring ateachraster laser scanlocation. Foreachobserved photon, wealso haveaccess tothetimeat whichitwasobserved, its characteristic emission spectrum, anditscharacteristic excited state lifetime. Thechallenge hereistousesmall numbers ofrandom, discrete, marked events suchasthese toperform accurate inference ontheunderlying phenomenon. Themarks associated witheachobserved event, while relevant toapplication-specific objectives, donotdirectly lead toimproved reconstruction capabilities inthese data-starved scenarios. Whenthenumberofobserved events isseverely limited, ourability toperform veryhigh-resolution orhighdimensional reconstruction canrapidly becomecritically impaired. Thisisoften counteracted byreducing thedimensionality oftheproblem inalinear, highly suboptimal mannerviabinning observations spatially and/or regardless of
BibTeX
@inproceedings{willett2007multiscaleintensityestimationfor,
  author = {Rebecca Willett},
  title = {{MULTISCALEINTENSITYESTIMATIONFOR MARKED POISSONPROCESSES}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2007},
  doi = {10.1109/icassp.2007.367070},
}

Loading related content...

Figure