The Frequency-Domain Effects of Stochastic Image Foveation in Superpixelating CamerasIntroduction and MotivationHalftoning by Classical ScreeningHalftoning by Error Diffusion [Floyd & Steinberg, 1976]Halftoning by Dithering with Blue Noise[Ulichney, 1988]What I Propose To DoReferencesThe FrequencyThe Frequency--Domain Effects of Domain Effects of Stochastic Image Foveation in Stochastic Image Foveation in SuperpixelatingSuperpixelatingCamerasCamerasThayne CoffmanThayne CoffmanEE381KEE381K--1414March 10, 2005March 10, 2005Introduction and MotivationIntroduction and MotivationSuperpixellatingSuperpixellatingcameras cameras [[McCarleyMcCarleyet al, 2004]et al, 2004]Images foveated directly on Images foveated directly on sensing plane by sharing pixel sensing plane by sharing pixel chargeschargesDesired resolution function Desired resolution function must be translated to binary must be translated to binary share/noshare/no--share control signalshare control signalCan achieve Can achieve ≥≥1000 1000 frfr/sec/secTranslation of control signal is Translation of control signal is similar to halftoningsimilar to halftoningWhich halftoning Which halftoning method would give the method would give the best ATR performance?best ATR performance?Halftoning by Classical ScreeningHalftoning by Classical ScreeningThe first widely used methodThe first widely used methodA point operation (very efficient)A point operation (very efficient)Pixel gray levels are Pixel gray levels are thresholdedthresholdedagainst a periodic dithering matrixagainst a periodic dithering matrixClustered dot matrices for printingClustered dot matrices for printingDispersed dot matrices for human Dispersed dot matrices for human consumptionconsumptionThe periodic dithering matrix can The periodic dithering matrix can introduce unpleasant visual artifactsintroduce unpleasant visual artifactsStandard dithering matrices were Standard dithering matrices were introduced in introduced in [Bayer, 1973]Clustered, 9 gray levelsClustered, 9 gray levelsUnclusteredUnclustered, 9 gray levels, 9 gray levelsDithered Dithered ‘‘LenaLena’’(dispersed dot)(dispersed dot)[Bayer, 1973]Halftoning by Error Diffusion Halftoning by Error Diffusion [Floyd & Steinberg, 1976][Floyd & Steinberg, 1976]■■7/167/163/163/165/165/161/161/16[some images from [some images from UlichneyUlichney, 1988], 1988]A neighborhood operation A neighborhood operation (more computation)(more computation)Quantization threshold fixed at 0.5Quantization threshold fixed at 0.5Feedback Feedback ““diffusesdiffuses””quantization quantization error by altering the gray values of error by altering the gray values of neighboring pixelsneighboring pixelsVisual quality is superior to screeningVisual quality is superior to screeningFeedback loopFeedback loopDiffusion Diffusion weightsweightsDithered Dithered ‘‘LenaLena’’Raster scanRaster scanSerpentine scanSerpentine scanHalftoning by Dithering with Blue NoiseHalftoning by Dithering with Blue Noise[[UlichneyUlichney, 1988], 1988]An extension of error An extension of error diffusiondiffusionDiffusion weights and/or Diffusion weights and/or locations are randomly locations are randomly ““perturbedperturbed””Result is isotropic highResult is isotropic high--frequency (frequency (““blueblue””) noise in ) noise in the dither pattern the dither pattern (which is good)(which is good)This reduces negative visual This reduces negative visual artifactsError diffusionError diffusionBlue noiseBlue noiseIdeal Ideal characteristicscharacteristicsartifactsWhat I Propose To DoWhat I Propose To DoPick some test images or video sequencesPick some test images or video sequencesPick the foveation points / desired resolution by handPick the foveation points / desired resolution by handTry out a few halftoning methods for control signal translationTry out a few halftoning methods for control signal translationDownload standard halftoning implementations Download standard halftoning implementations [[MongaMongaet al 2002]et al 2002]Attempt to make my own stochastic approachAttempt to make my own stochastic approachSimulate the charge sharing behavior of the camera in softwareSimulate the charge sharing behavior of the camera in softwareEvaluate the quality of the foveated images via some subset ofEvaluate the quality of the foveated images via some subset ofMSE in frequency domainMSE in frequency domainSNR, PSNR, WSNRSNR, PSNR, WSNRFoveated Wavelet Quality Index (FWQI) Foveated Wavelet Quality Index (FWQI) [Wang et al 2001][Wang et al 2001]UlichneyUlichney’’ssanisotropy and frequency metrics on power spectraanisotropy and frequency metrics on power spectraReferencesReferencesB.E. Bayer, B.E. Bayer, ““An optimum method for two level rendition of continuousAn optimum method for two level rendition of continuous--tone tone pictures,pictures,””Proc. IEEE Int. Conf. on Communications, Conf. Rec.Proc. IEEE Int. Conf. on Communications, Conf. Rec., pp. (26, pp. (26--11)11)--(26(26--15), 15), 1973.1973.R. Floyd, L. Steinberg, R. Floyd, L. Steinberg, ““An adaptive algorithm for spatial grayscale,An adaptive algorithm for spatial grayscale,””Proc. SIDProc. SID’’7676, , pp. 75pp. 75--77, 1976. 77, 1976. P. P. McCarleyMcCarley, M. Massie, J.P. , M. Massie, J.P. CurzanCurzan, , ““Large format variable spatial acuity superpixel Large format variable spatial acuity superpixel imaging: visible and infrared systems applications,imaging: visible and infrared systems applications,””Proc. SPIE, Infrared Technology and Proc. SPIE, Infrared Technology and Applications Applications XXXXXX[sic[sic.], vol. 5406, pp. 361.], vol. 5406, pp. 361--369, Aug 2004. 369, Aug 2004. V. V. MongaMonga, N. , N. DameraDamera--VenkataVenkata, B. Evans, , B. Evans, Halftoning Toolbox for MatlabHalftoning Toolbox for Matlab..Version 1.1 Version 1.1 released November 7, 2002. Available online at released November 7, 2002. Available online at http://http://www.ece.utexas.edu/~bevans/projects/halftoningwww.ece.utexas.edu/~bevans/projects/halftoning//. . R.A. R.A. UlichneyUlichney, , ““Dithering with blue noise,Dithering with blue noise,””Proc. IEEEProc. IEEE, vol. 76, pp. 56, vol. 76, pp. 56--79,
View Full Document