Tag Archives: noise

The Richardson-Lucy Algorithm

Deconvolution by the Richardson-Lucy algorithm is achieved by minimizing the convex loss function derived in the last article

(1)   \begin{equation*} J(O) = \sum \bigg (O**PSF - I\cdot ln(O**PSF) \bigg) \end{equation*}

with

  • J, the scalar quantity to minimize, function of ideal image O(x,y)
  • I(x,y), linear captured image intensity laid out in M rows and N columns, corrupted by Poisson noise and blurring by the PSF
  • PSF(x,y), the known two-dimensional Point Spread Function that should be deconvolved out of I
  • O(x,y), the output image resulting from deconvolution, ideally without shot noise and blurring introduced by the PSF
  • **   two-dimensional convolution
  • \cdot   element-wise product
  • ln, element-wise natural logarithm

In what follows indices x and y, from zero to M-1 and N-1 respectively, are dropped for readability.  Articles about algorithms are by definition dry so continue at your own peril.

So, given captured raw image I blurred by known function PSF, how do we find the minimum value of J yielding the deconvolved image O that we are after?

Continue reading The Richardson-Lucy Algorithm

Elements of Richardson-Lucy Deconvolution

We have seen that deconvolution by naive division in the frequency domain only works in ideal conditions not typically found in normal photographic settings, in part because of shot noise inherent in light from the scene. Half a century ago William Richardson (1972)[1] and Leon Lucy (1974)[2] independently came up with a better way to deconvolve blurring introduced by an imaging system in the presence of shot noise. Continue reading Elements of Richardson-Lucy Deconvolution

Sub LSB Quantization

This article is a little esoteric so one may want to skip it unless one is interested in the underlying mechanisms that cause quantization error as photographic signal and noise approach the darkest levels of acceptable dynamic range in our digital cameras: one least significant bit in the raw data.  We will use our simplified camera model and deal with Poissonian Signal and Gaussian Read Noise separately – then attempt to bring them together.

Continue reading Sub LSB Quantization

Photographic Sensor Simulation

Physicists and mathematicians over the last few centuries have spent a lot of their time studying light and electrons, the key ingredients of digital photography.  In so doing they have left us with a wealth of theories to explain their behavior in nature and in our equipment.  In this article I will describe how to simulate the information generated by a uniformly illuminated imaging system using open source Octave (or equivalently Matlab) utilizing some of these theories.

Since as you will see the simulations are incredibly (to me) accurate, understanding how the simulator works goes a long way in explaining the inner workings of a digital sensor at its lowest levels; and simulated data can be used to further our understanding of photographic science without having to run down the shutter count of our favorite SLRs.  This approach is usually referred to as Monte Carlo simulation.

Continue reading Photographic Sensor Simulation

Information Theory for Photographers

Ever since Einstein we’ve been able to say that humans ‘see’ because information about the scene is carried to the eyes by photons reflected by it.  So when we talk about Information in photography we are referring to information about the energy and distribution of photons arriving from the scene.   The more complete this information, the better we ‘see’.  No photons = no information = no see; few photons = little information = see poorly = poor IQ; more photons = more information = see better = better IQ.

Sensors in digital cameras work similarly, their output ideally being the energy and location of every photon incident on them during Exposure. That’s the full information ideally required to recreate an exact image of the original scene for the human visual system, no more and no less. In practice however we lose some of this information along the way during sensing, so we need to settle for approximate location and energy – in the form of photoelectron counts by pixels of finite area, often correlated to a color filter array.

Continue reading Information Theory for Photographers

Determining Sensor IQ Metrics: RN, FWC, PRNU, DR, gain – 2

There are several ways to extract Sensor IQ metrics like read noise, Full Well Count, PRNU, Dynamic Range and others from mean and standard deviation statistics obtained from a uniform patch in a camera’s raw file.  In the last post we saw how to do it by using such parameters to make observed data match the measured SNR curve.  In this one we will achieve the same objective by fitting mean and  standard deviation data.  Since the measured data is identical, if the fit is good so should be the results.

Sensor Metrics from Measured Mean and Standard Deviation in DN

Continue reading Determining Sensor IQ Metrics: RN, FWC, PRNU, DR, gain – 2