Keigo Hirakawa

Biography

Keigo Hirakawa received the B.S.E. degree in electrical engineering from Princeton University, Princeton, NJ, in 2000 with high honors, the M.S. and Ph.D. degrees in electrical and computer engineering from Cornell University, Ithaca, NY, in 2003 and 2005, respectively, and the M.M. degree in Jazz Performance from the New England Conservatory of Music, Boston, MA, in 2006 with high honors. He is currently with the School of Engineering and Applied Sciences and the Department of Statistics at Harvard University.

Hirakawa's cross-disciplinary research aims to bridge statistics and signal processing where neither of them can offer adequate solution alone. His work focuses on harmonic analysis of signals with a particular emphasis on statistical modeling and hardware interactions. Examples of sampling, missing data, and estimation problems of such flavor appear in medical/consumer imaging, communication, computer vision, and the like. Robust statistical modeling and processing of signal data can therefore be viewed as an important testbed for making a high impact in a number of complex and emerging signal processing problems.

Research Interests

Hirakawa's contributions are most well known for the work related to color imaging and digital cameras. His research in optimal color image data acquisition and display has enjoyed many recognitions---including the honor of delivering the keynote address at IS&T CGIV in 2008 and the Docomo USA Labs Most Innovative Paper Award at IEEE ICIP 2007. His widely popular AHD demosaicking algorithm is by now the de facto standard for most open-sourced raw image converters, and he has a contract to develop a novel denoising+demosaicking solution with Sony. He has previously been an ASIC engineer and principal image scientist for the camera division of Hewlett-Packard/Agilent Technologies, and his past and current collaborations with camera/display manufacturers include Sony, Micron, Texas Instruments, AZ Electronic Materials, and NEC.

Hirakawa has made key advances in DSP and statistics also. A novel work on wavelet-based Bayesian Poisson rate estimation has been applied to various forms of "count data," including MRI (# spins), image sensors (# photons), X-ray sensors (# photons), and network reliability analysis (# packets). His recent research yields a proof that multiresolution analysis admits a canonical time-frequency aliasing structure analogous to the traditional sampling theory in the Fourier domain. He has also shown a pragmatic approach to rigorously applying Bayesian hierarchical model to wavelet coefficients when no transform coefficients are observable due to severe undersampling.

Hirakawa's interests extend to basic science as well. One notorious problem that remains unsolved in color science is the detection of the color of the light source illuminating a natural scene. Hirakawa made a fundamental observation in spatio-spectral representation that admits clean disambiguation between the reflectance and the light source. This approach is recently extended to a new spectral "super-resolution" method that stems from spatio-spectral regression, with applications to hyperspectral imaging. These discoveries represent progress in computer vision, computer graphics, and remote sensing, where accurate measurements of reflectance values are needed.

Professional Activities

Hirakawa is an author of two book chapters and many journal and conference presentations. He has received recognitions from IEEE, IS&T, Docomo USA Labs, and Lockheed Martin and has co-authored multiple successful grant applications. He maintains active collaborations across disciplines (statistics, computer vision, DSP) and universities (Harvard, Cornell, WVU, UVA, UCSD). As an educator, he is among the favorites of students at Princeton, Cornell, and Harvard. Besides lecturing introductory statistics course, he chairs committees to evaluate teaching resources at Harvard and to train new graduate students to become teaching assistants. He currently mentors two Ph.D students at Harvard. He is frequently seen leading seminars at high tech companies in Silicon Valley.

Book Chapters

K.H., "Color Filter Array Image Analysis for Joint Denoising and Demosaicking," in Single-Sensor Imaging: Methods and Applications for Digital Cameras, ed. R. Lukac, CRC Press, 2008. [pdf]
K.H., P.J. Wolfe, "Spatio-Spectral Sampling and Color Filter Array Design," in Single-Sensor Imaging: Methods and Applications for Digital Cameras, ed. R. Lukac, CRC Press, 2008. [pdf]

Journal Papers

K.H., P.J. Wolfe, "Optimal Color Filter Array Design by Spatio-Spectral Sampling," IEEE TIP, October 2008.[pdf]
K.H., T.W. Parks, "Image Denoising using Total Least Squares," IEEE TIP September 2006. [pdf]
K.H., T.W. Parks, "Joint Demosaicing and Denoising," IEEE TIP August 2006. [pdf]
K.H., T.W. Parks, "Adaptive Homogeneity-Directed Demosaicing Algorithm," IEEE TIP, March 2005. [pdf]
A. Chakrabarti, K.H., T. Zickler, "Computational Color Constancy with Spatial Correlations," IEEE PAMI, under review. [pdf]
K.H., P.J. Wolfe, "Wavelet- and Filterbank-Based Poisson Intensity Estimation Using the Skellam Distribution," under review by sponsor, to be submitted to IEEE TIT. [pdf]
K.H., F. Baqai, P.J. Wolfe, "Poisson Intensity Estimation for Image Denoising: A Comparative Study," under review by sponsor, to be submitted to IEEE TIP. [pdf]
K.H., P.J. Wolfe, "Reverse-Ordered and Permuted Subband Structures in Filterbank Transforms: Theory and Application," in preparation, to be submitted to IEEE TIT. [pdf]
K.H., X.-L. Meng, "Empirical Partial Bayesian Hierarchical Modeling of Image Wavelet Coefficients with Missing Data," in preparation, to be submitted to IEEE TIP. [pdf]

Conference Presentations

K.H., P.J. Wolfe, "Efficient Multivariate Skellam Shrinkage for Denoising Photon-Limited Image Data: An Empirical Bayes Approach," IEEE ICIP 2009, under review. [pdf]
K.H., P.J. Wolfe, "SkellamShrink: Poisson Intensity Estimation for Vector-Valued Data," IEEE ICASSP 2009. [pdf]
K.H., F. Baqai, P.J. Wolfe, "Wavelet-Based Poisson Rate Estimation Using the Skellam Distribution," SPIE EI/CIC, 2009. [pdf]
K.H., "Spatio-Spectral Sampling in Multispectral Imaging," IS&T CGIV, 2008 (keynote address).
K.H., P.J. Wolfe, T. Nguyen, "Color Imaging Pipeline for Digital Still and Video Cameras," IEEE ICIP, 2008 (tutorial session, website).
K.H., "Cross-Talk Explained," IEEE ICIP, 2008. [pdf]
A. Chakrabarti, K.H., T. Zickler, "Color Constancy with Spatial Correlations," Workshop: Perception of Material Properties in 3D Scenes, U Penn, 2008. [pdf]
A. Chakrabarti, K.H., T. Zickler, "Color Constancy Beyond Bag Of Pixels," IEEE CVPR, 2008. [pdf]
K.H., P.J. Wolfe, "Advancing The Digital Camera Pipeline For Mobile Multimedia: Key Challenges From a Signal processing Perspective," IEEE ICASSP, 2008 (special session: invited). [pdf]
A. Chakrabarti, K.H., "Effective Separation of Sparse and Non-Sparse Image Features For Denoising," IEEE ICASSP, 2008. [pdf]
K.H., "Fourier and Filterbank Analysis of Signal-Dependent Noise," IEEE ICASSP, 2008. [pdf]
K.H., P.J. Wolfe, "Second Generation CFA and Demosaicking Design," SPIE EI/VCIP, 2008 (invited). [pdf]
K.H., P.J. Wolfe, "Spatio-Spectral Color Filter Array for Enhanced Image Fidelity," IEEE ICIP, 2007 (paper award). [pdf]
K.H., P.J. Wolfe, "Fourier Domain Display Color Filter Array Design for Enhanced Image Fidelity," IEEE ICIP, 2007. [pdf]
K.H., "Signal-Dependent Noise Characterization in Haar Filterbank Representation," SPIE O&P/Wavelets, 2007 (invited). [pdf]
K.H., X.-L. Meng, P.J. Wolfe, "A Framework for Wavelet-Based Analysis and Processing of Color Filter Array Images with Applications to Denoising and Demosaicing," IEEE ICASSP, 2007. [pdf]
K.H., X.-L. Meng, "An Empirical Bayes EM-Wavelet Unification for Simultaneous Denoising, Interpolation, and/or Demosaicing," IEEE ICIP, 2006. [pdf]
K.H., T.W. Parks, "Image Denoising for Signal-Dependent Noise" IEEE ICASSP, 2005. [pdf]
K.H., T.W. Parks, "Joint Demosaicing and Denoising," IEEE ICIP, 2005. [pdf]
K.H., T.W. Parks, "Chromatic Adaptation and White-Balance Problem", IEEE ICIP, 2005. [pdf]
K.H., T.W. Parks, "Adaptive Homogeneity-Directed Demosaicing Algorithm," IEEE ICIP, 2003. [pdf]
K.H., T.W. Parks, "Tone-Curve Filtering," IEEE WNYIP Workshop, 2003. [pdf]
K.H., H. Igura, "Parallel DSP Task-Management Algorithms," IEICE Info. Sys. Soc., 1998. [pdf]

Patents

K.H., P.J. Wolfe, "A Novel Color Filter Array Design," submitted 2006. (link)
K.H., X.-L. Meng, P.J. Wolfe, "Wavelet-Based Denoising and Demosaicing," submitted 2006. (link)
K.H., "System and Method for Cross-talk Correction," submitted 2008.
A. Chakrabarti, K.H., T. Zickler, "System and Method for Color Constancy" submitted 2008.

Time-Frequency Analysis of Subsampled Signal

This work shows that reversing the ordering of filterbanks/wavelets (from high to low frequency) results in exact frequency modulation of the input signal, enabling an exact and efficient time-frequency analysis of subsampled signals. We provide a proof that multiresolution analysis admits a canonical time-frequency aliasing structure analogous to the well understood Fourier sampling theory. Applications include representation, analysis, and processing of subsampled data, including transform domain interpolation, denoising, and coding.

K.H., P.J. Wolfe, "Reverse-Ordered and Permutation Scale Structures in Filterbank Transforms," in preparation, to be submitted to IEEE TIT. [pdf]
K. Hirakawa, X.-L. Meng, P.J. Wolfe, "A Framework for Wavelet-Based Analysis and Processing of Color Filter Array Images with Applications to Denoising and Demosaicing," ICASSP 2007. [pdf]
K.H., "Color Filter Array Image Analysis for Joint Denoising and Demosaicking," in Single-Sensor Imaging: Methods and Applications for Digital Cameras, ed. R. Lukac, CRC Press, 2008. [pdf]

Permutative Filterbank Transform

In this work, we show that a "pointwise" multiplication of two filterbank/wavelet analysis functions can lead to a third analysis function. Certain filterbank transforms therefore exhibit a self-permutative structure which can be useful in representing, analyzing, and processing a number of phenomenons that are otherwise to complicated to characterize in the time-frequency domain---including non-uniform/asynchronous sampling, signal dependent noise, and Poisson process.

K.H., P.J. Wolfe, "Reverse-Ordered and Permutation Scale Structures in Filterbank Transforms," in preparation, to be submitted to IEEE TIT. [pdf]
K.H., "Fourier and Filterbank Analysis of Signal-Dependent Noise," IEEE ICASSP, 2008. [pdf]
K.H., "Signal-Dependent Noise Characterization in Haar Filterbank Representation," SPIE O&P/Wavelets, 2007. [pdf]

Wavelet-Based Processing with Missing Data

Every wavelet coefficient is a linear combination of the signal samples. Thus when samples are missing, wavelet coefficients are no longer observable because it is not possible to perform a linear transformation. The coupling of EM algorithm with the Bayesian hierarchical modeling of wavelet coefficients, however, offers a statistically principled, flexible, and pragmatic approach to wavelet-based signal processing combined with proper missing data treatment.

K.H., X.-L. Meng, "Empirical Partial Bayesian Hierarchical Modeling of Image Wavelet Coefficients with Missing Data," in preparation, to be submitted to IEEE TIP. [pdf]
K.H., X.-L. Meng, "An Empirical Bayes EM-Wavelet Unification for Simultaneous Denoising, Interpolation, and/or Demosaicing," IEEE ICIP, 2006. [pdf]

Signal Dependent Noise

Owing to the stochastic nature of discrete processes such as photon, spin, or network packet counts, real-world data measurements often exhibit heteroscedastic behavior. In particular, time series components and other measurements may frequently be assumed to be non-iid Poisson random variables, whose rate parameter is proportional to the underlying signal of interest. In this work, we show that certain wavelet and filterbank transform coefficients corresponding to vector-valued measurements of this type are distributed as sums and differences of independent Poisson counts. While exact estimates rarely admit analytical forms, we derive rate estimators under both frequentist and Bayes models.

K.H., P.J. Wolfe, "Wavelet- and Filterbank-Based Poisson Intensity Estimation Using the Skellam Distribution," under review by sponsor, to be submitted to IEEE TIT. [pdf]
K.H., F. Baqai, P.J. Wolfe, "Poisson Intensity Estimation for Image Denoising: A Comparative Study," under review by sponsor, to be submitted to IEEE TIP. [pdf]
K.H., P.J. Wolfe, "SkellamShrink: Poisson Intensity Estimation for Vector-Valued Data," IEEE ICASSP 2009. [pdf]
K.H., F. Baqai, P.J. Wolfe, "Wavelet-Based Poisson Rate Estimation Using the Skellam Distribution," SPIE EI/CIC, 2009. [pdf]
K.H., "Fourier and Filterbank Analysis of Signal-Dependent Noise," IEEE ICASSP, 2008. [pdf]
K.H., "Signal-Dependent Noise Characterization in Haar Filterbank Representation," SPIE O&P/Wavelets, 2007 (invited). [pdf]

Total Least Squares Estimate

An alternative to least squares approach, total least square methods yield a more precise regression when the underlying model as well as the observations are noisy. We use TLS estimates for prediction, denoising, and interpolation for autoregressive and linear models.

K.H., T.W. Parks, "Image Denoising using Total Least Squares," IEEE TIP September 2006. [pdf]

Nuclear Medicine: MRI, CT

Poisson process is inherent in nuclear medicine because the signal is proportional to proton spin and photon counts in MRI and CT imaging. However, Poisson intensity estimation is a long outstanding problem despite its grave importance. For this reason, the overwhelming majority of existing denoising techniques oversimplify the real-world noise models in order to circumvent the complex interplay between the noise, the signal, and the transform. However, the abelian group structure in the canonical transform operators yields mechanism for characterizing the heteroscedasticity in the transform domain. The comprehensive Poisson intensity estimation strategy based on an accurate "noise model" not only affords quantifiable rate of performance but also provides concrete steps toward merging the representation of real noise with the transform-based signal processing strategies. An explicit treatment Poisson process that arise in nuclear medicine is amenable to efficient denoising schemes with clear computational and analytical advantages over the existing alternatives, which are by and large ad-hoc.

K.H., P.J. Wolfe, "Wavelet- and Filterbank-Based Poisson Intensity Estimation Using the Skellam Distribution," under review by sponsor, to be submitted to IEEE TIT. [pdf]
K.H., P.J. Wolfe, "SkellamShrink: Poisson Intensity Estimation for Vector-Valued Data," IEEE ICASSP 2009. [pdf]
K.H., F. Baqai, P.J. Wolfe, "Wavelet-Based Poisson Rate Estimation Using the Skellam Distribution," SPIE EI/CIC, 2009. [pdf]
K.H., "Fourier and Filterbank Analysis of Signal-Dependent Noise," IEEE ICASSP, 2008. [pdf]
K.H., "Signal-Dependent Noise Characterization in Haar Filterbank Representation," SPIE O&P/Wavelets, 2007 (invited). [pdf]

Spatio-Spectral Sampling: Color Filter Array

In this work, we present a solution to the problem of optimal color filter array design, from the point of view of jointly minimizing spatial and spectral information loss. We show (1) how to quantify the information loss associated with color image acquisition; (2) how to analyze the fundamental limitations to subsequent processing imposed by current hardware; (3) how to design new color filter arrays that minimize these losses and limitations; and (4) how to achieve corresponding gains in image quality through fast linear reconstruction algorithms. The methodologies we present define a new paradigm for capture, processing, and display of color images that will significantly reduce hardware complexity in applications such as digital still and video cameras, while at the same time improving output color image quality.

K.H., P.J. Wolfe, "Optimal Color Filter Array Design by Spatio-Spectral Sampling," IEEE TIP, October 2008.[pdf]
K.H., P.J. Wolfe, "Spatio-Spectral Sampling and Color Filter Array Design," in Single-Sensor Imaging: Methods and Applications for Digital Cameras, ed. R. Lukac, CRC Press, 2008. [pdf]
K.H., "Spatio-Spectral Sampling in Multispectral Imaging," IS&T CGIV, 2008 (keynote address).
K.H., P.J. Wolfe, "Spatio-Spectral Color Filter Array for Enhanced Image Fidelity," IEEE ICIP, 2007 (paper award). [pdf]

Spatio-Spectral Reconstruction: Demosaicking, Denoising, Cross-Talk

Reconstruction and processing of subsampled color image, such as that of the raw image sensor data taken under color filter array, are generally ill-posed inverse problems which require additional assumptions about the structure in the signal. If the spatio-spectral sampling is designed well, however, the complexity of demosaicking can be reduced to the order of bilinear interpolation, even while yielding image quality unmatched by any state-of-the-art demosaicking method. We rigorously characterize and correct cross-talk (photon and electron leakage) artifacts, and introduce novel solutions for simultaneous denoising and demosaicking.

K.H., "Cross-Talk Explained," IEEE ICIP, 2008. [pdf]
K.H., P.J. Wolfe, "Second Generation CFA and Demosaicking Design," SPIE EI/VCIP, 2008 (invited). [pdf]
K.H., "Color Filter Array Image Analysis for Joint Denoising and Demosaicking," in Single-Sensor Imaging: Methods and Applications for Digital Cameras, ed. R. Lukac, CRC Press, 2008. [pdf]
K.H., X.-L. Meng, P.J. Wolfe, "A Framework for Wavelet-Based Analysis and Processing of Color Filter Array Images with Applications to Denoising and Demosaicing," IEEE ICASSP, 2007. [pdf]
K.H., X.-L. Meng, "An Empirical Bayes EM-Wavelet Unification for Simultaneous Denoising, Interpolation, and/or Demosaicing," IEEE ICIP, 2006. [pdf]
K.H., T.W. Parks, "Joint Demosaicing and Denoising," IEEE TIP August 2006. [pdf]
K.H., T.W. Parks, "Joint Demosaicing and Denoising," IEEE ICIP, 2005. [pdf]
K.H., T.W. Parks, "Adaptive Homogeneity-Directed Demosaicing Algorithm," IEEE TIP, March 2005. [pdf]
K.H., T.W. Parks, "Adaptive Homogeneity-Directed Demosaicing Algorithm," IEEE ICIP, 2003. [pdf]

Spatio-Spectral Display: Display Device

The color filter array implemented in display devices such as plasma and LCD pose the "dual" problem to what we study in color image acquisition. Aliasing structure caused by color filter array in display devices impose fundamental limitation to the human visual system. We overcome this by reinterpreting the interaction between the display device and the human visual system as amplitude demodulation, and maximizing the throughput of visual information communicated to the end-user. This idea can also be leveraged for image projectors and 3D displays.

K.H., P.J. Wolfe, "Fourier Domain Display Color Filter Array Design for Enhanced Image Fidelity," IEEE ICIP, 2007. [pdf]

Spatio-Spectral Analysis: Illuminant, Reflectance, Color

Estimating the spectral distribution of scene illuminant and surface reflectance often plays a central role in computer vision, graphics, and remote sensing. While these problems have received significant attention, the methods that exist do not maximally leverage spatial dependencies between pixels. Indeed, most methods treat the observed color (or its spatial derivative) at each pixel independently of its neighbors. We propose an alternative approach to illuminant and reflectance estimation---one that employs an explicit statistical model to capture the spatial dependencies between pixels induced by the surfaces they observe.

K.H., "Spatio-Spectral Sampling in Multispectral Imaging," IS&T CGIV, 2008 (keynote address).
A. Chakrabarti, K.H., T. Zickler, "Computational Color Constancy with Spatial Correlations," IEEE PAMI, under review. [pdf]
A. Chakrabarti, K.H., T. Zickler, "Color Constancy Beyond Bag Of Pixels," IEEE CVPR, 2008. [pdf]
A. Chakrabarti, K.H., T. Zickler, "Color Constancy with Spatial Correlations," Perception of Material Properties in 3D Scenes, 2008. (workshop) [pdf]
K.H., T.W. Parks, "Chromatic Adaptation and White-Balance Problem", IEEE ICIP, 2005. [pdf]

Digital Camera

What is the state-of-the-art of digital cameras? What are the foreseeable problems in the future? Find out more here.

K.H., P.J. Wolfe, T. Nguyen, "Color Imaging Pipeline for Digital Still and Video Cameras," IEEE ICIP, 2008 (tutorial session).
K.H., "Cross-Talk Explained," IEEE ICIP, 2008. [pdf]
K.H., P.J. Wolfe, "Advancing The Digital Camera Pipeline For Mobile Multimedia: Key Challenges From a Signal processing Perspective," IEEE ICASSP, 2008 (special session: invited). [pdf]

Total Least Squares Estimate

This is a new approach to image denoising using an image patch modeling technique. It is capable of removing additive, multiplicative, impulsive, and mixed noise--ideal for CMOS and CCD image sensor noise. The results are comparable to or better than the state-of-the-art denoising algorithms for signal-independent noise, and better for signal-dependent noise.

K.H., T.W. Parks, "Image Denoising using Total Least Squares," IEEE TIP September 2006. [pdf]
K.H., T.W. Parks, "Image Denoising for Signal-Dependent Noise" IEEE ICASSP, 2005. [pdf]

Sparsity and Image Denoising

Over-complete representations of images such as undecimated wavelets have enjoyed immense popularity in recent years. Though they are efficient for modeling singularities and edges, natural images also consist of textures that are difficult to capture with any canonical transformation. In this work, we develop a new modeling strategy with a rigorous treatment of textured regions. Using principal components analysis as an approximate classifier for edges and textures, we partition an image into compressible and incompressible regions---with corresponding models matching their behaviors.

A. Chakrabarti, K.H., "Effective Separation of Sparse and Non-Sparse Image Features For Denoising," IEEE ICASSP, 2008. [pdf]

Signal Dependent Noise

Signal-dependent noise removal is a long outstanding problem despite its grave importance. For this reason, the overwhelming majority of existing denoising techniques oversimplify the real-world noise models in order to circumvent the complex interplay between the noise, the signal, and the transform. However, the abelian group structure in the canonical transform operators is the key mechanism for characterizing the signal-dependent noise in the transform domain. The comprehensive signal denoising strategy based on an accurate noise model not only affords quantifiable rate of performance but also provides concrete steps toward merging the representation of real noise with the transform-based signal processing strategies. An explicit treatment of noise that arise in biomedical/consumer imaging and astronomy is amenable to efficient denoising schemes with clear computational and analytical advantages over the existing alternatives, which are by and large ad-hoc.

K.H., P.J. Wolfe, "Wavelet- and Filterbank-Based Poisson Intensity Estimation Using the Skellam Distribution," under review by sponsor, to be submitted to IEEE TIT. [pdf]
K.H., P.J. Wolfe, "SkellamShrink: Poisson Intensity Estimation for Vector-Valued Data," IEEE ICASSP 2009, under review. [pdf]
K.H., F. Baqai, P.J. Wolfe, "Wavelet-Based Poisson Rate Estimation Using the Skellam Distribution," SPIE EI/CIC, 2009. [pdf]
K.H., "Fourier and Filterbank Analysis of Signal-Dependent Noise," IEEE ICASSP, 2008. [pdf]
K.H., "Signal-Dependent Noise Characterization in Haar Filterbank Representation," SPIE O&P/Wavelets, 2007 (invited). [pdf]

Poisson Intensity Estimation

K.H., P.J. Wolfe, "Wavelet- and Filterbank-Based Poisson Intensity Estimation Using the Skellam Distribution," under review by sponsor, to be submitted to IEEE TIT. [pdf]
K.H., P.J. Wolfe, "SkellamShrink: Poisson Intensity Estimation for Vector-Valued Data," IEEE ICASSP 2009, under review. [pdf]
K.H., F. Baqai, P.J. Wolfe, "Wavelet-Based Poisson Rate Estimation Using the Skellam Distribution," SPIE EI/CIC, 2009. [pdf]

Signal Dependent Noise

K.H., "Fourier and Filterbank Analysis of Signal-Dependent Noise," IEEE ICASSP, 2008. [pdf]
K.H., "Signal-Dependent Noise Characterization in Haar Filterbank Representation," SPIE O&P/Wavelets, 2007 (invited). [pdf]

Sparsity and Image Denoising

Over-complete representations of images such as undecimated wavelets have enjoyed immense popularity in recent years. Though they are efficient for modeling singularities and edges, natural images also consist of textures that are difficult to capture with any canonical transformation. In this work, we develop a new Bayesian modeling strategy with a rigorous treatment of textured regions. Using principal components analysis as an approximate classifier for edges and textures, we partition an image into compressible and incompressible regions---with corresponding models matching their behaviors.

A. Chakrabarti, K.H., "Effective Separation of Sparse and Non-Sparse Image Features For Denoising," IEEE ICASSP, 2008. [pdf]

Reconstruction of Asynchronously Sampled Signal

The attraction to working with an asynchronous sampling is that the samples arrive �as necessary�---a level-crossing A/D converter is one such example. However, reconstruction from irregularly spaced samples using traditional signal processing techniques is extremely difficult. A quantifiably rigorous solution to this problem is to appeal to the notion of sparsity. When combined with the permutation filterbank techniques we see a unified approach to designing a new circuitry with a fast sample rate.

Illuminant and Reflectance

K.H., "Spatio-Spectral Sampling in Multispectral Imaging," IS&T CGIV, 2008 (keynote address).
A. Chakrabarti, K.H., T. Zickler, "Computational Color Constancy with Spatial Correlations," in preparation, to be submitted to IEEE PAMI. [pdf]
A. Chakrabarti, K.H., T. Zickler, "Color Constancy Beyond Bag Of Pixels," IEEE CVPR, 2008. [pdf]
A. Chakrabarti, K.H., T. Zickler, "Color Constancy with Spatial Correlations," Perception of Material Properties in 3D Scenes, 2008. (workshop) [pdf]
K.H., T.W. Parks, "Chromatic Adaptation and White-Balance Problem", IEEE ICIP, 2005. [pdf]

Keigo Hirakawa - Statistical Signal Processing Research

News

bio cv publications jazz

Biography

Research Interests

Professional Activities

Book Chapters

Journal Papers

Conference Presentations

Patents

Signal Processing

wavelets denoising asynchronous

Time-Frequency Analysis of Subsampled Signal

Permutative Filterbank Transform

Wavelet-Based Processing with Missing Data

Reconstruction of Asynchronously Sampled Signal

Signal Dependent Noise

Total Least Squares Estimate

Image Processing

biomedical spatio-spectral denoising

Nuclear Medicine: MRI, CT

Spatio-Spectral Sampling: Color Filter Array

Spatio-Spectral Reconstruction: Demosaicking, Denoising, Cross-Talk

Spatio-Spectral Display: Display Device

Spatio-Spectral Analysis: Illuminant, Reflectance, Color

Digital Camera

Total Least Squares Estimate

Sparsity and Image Denoising

Signal Dependent Noise

Statistics and Modeling

missing data poisson sparsity

Signal Processing Techniques with Missing Data

Poisson Intensity Estimation

Signal Dependent Noise

Sparsity and Image Denoising

Reconstruction of Asynchronously Sampled Signal

Sciences

color science

Illuminant and Reflectance

Software

Color Constancy/White Balance CFA design TLS denoising TLS denoising+demosaicking AHD demosaicking

Color Constancy/White Balance

CFA design

TLS denoising

TLS denoising+demosaicking

AHD demosaicking