Relative vs. Absolute Imaging in Electrical Resistance Tomography (ERT)

TL;DR

Relative (difference) imaging reconstructs the change in conductivity between two states — two data frames are subtracted before inversion. It is robust and artifact-light, making it the default choice for industrial process monitoring. Absolute imaging reconstructs the absolute conductivity distribution at a single instant from one data frame. It needs no reference state and yields absolute values, but is sensitive to sensor modeling errors. In engineering practice, the mainstream approach is to combine both: absolute imaging to calibrate the baseline, relative imaging to track changes.

Two Routes to the Same Inverse Problem

The Mathematical Essence of ERT Imaging

In Electrical Resistance Tomography (ERT), current is injected through boundary electrodes, voltages are measured on the remaining electrodes, and the internal conductivity distribution is inferred from these boundary voltages.

Forward problem (known distribution → voltages):
  Conductivity σ(x, y)  ──F──▶  Boundary voltages V
  The forward model F is described by the complete
  electrode model (CEM), including electrode geometry,
  contact impedance, and insulation boundary conditions

Inverse problem (known voltages → distribution):
  Boundary voltages V  ──?──▶  Conductivity σ(x, y)
  σ̂ = argmin { ‖F(σ) − V‖² + λ·R(σ) }

This inverse problem has two well-known difficulties:

1. Nonlinearity: voltage depends nonlinearly on conductivity (a conductivity change simultaneously redistributes the current lines inside the field);
2. Ill-posedness: boundary voltage sensitivity to internal conductivity changes decays sharply with distance, strongly amplifying errors in the central region.

Facing the same difficult problem, engineering practice offers two fundamentally different “questions to ask,” corresponding to two technical routes:

Route 1 (relative imaging): "Relative to a reference state, what changed?"
  Input:  ΔV = V₂ − V₁ (difference of two data frames)
  Output: Δσ̂ (conductivity change distribution)

Route 2 (absolute imaging): "What is the conductivity distribution right now?"
  Input:  V (a single data frame)
  Output: σ̂ (absolute conductivity distribution)

Which question you choose to ask determines your data acquisition scheme, reconstruction algorithm, requirements on sensor accuracy, and — ultimately — what you can read out of the image.

Relative (Difference) Imaging: The Change Is the Object

Basic Principle

Relative imaging does not invert for the absolute distribution. Instead, it inverts for the difference between two states:

ΔV = V(t₂) − V(t₁)          ← subtraction in the data domain
Δσ̂ ≈ −J† · ΔV               ← linearized inversion

J:  sensitivity (Jacobian) matrix at the reference state t₁
J†: generalized inverse (used together with regularization)

Because ΔV usually corresponds to a small perturbation, the linearized approximation of the voltage–conductivity relationship holds well near the reference state, and the inverse problem is noticeably better conditioned than in absolute imaging.

Three Common Difference Schemes

Time-difference imaging:
├─ Reference frame: measured data at t₁ (empty pipe / full pipe / initial state)
├─ Characteristics: most common; most thorough error cancellation
└─ Requires: access to a stable reference state

Frequency-difference imaging:
├─ Reference frame: data at another excitation frequency, same instant
├─ Characteristics: for targets that change too fast to capture a reference frame
└─ Requires: exploitable frequency dependence of the target's electrical properties

Reference-object / normalized difference:
├─ Reference frame: simulated or measured data of a known phantom or uniform medium
├─ Characteristics: intermediate; often combined with normalized sensitivity
└─ Requires: how well the reference state matches the real one

Why Industrial Sites Prefer Relative Imaging

1. Common-mode errors cancel automatically. Inconsistency and drift of electrode contact impedance, electrode positioning errors, modeling errors of vessel geometry, temperature drift of the excitation amplitude — these errors are roughly the same in both frames and largely vanish after subtraction. The remaining ΔV carries almost exclusively the information about “what changed inside.”

2. The inverse problem is better conditioned. Linearization error is small, regularization techniques are mature, reconstructed images show markedly fewer artifacts, and the “perceived resolution” is higher.

3. Real-time performance. The sensitivity matrix can be computed once before operation; online reconstruction reduces to a single linear step, easily reaching high frame rates for continuous process monitoring.

What Relative Imaging Cannot See

• Changes only, no absolute values: the image answers “where did it change, and by how much (relative)” — not “what is the conductivity now”;
• Dependence on reference-frame quality and freshness: a poor reference frame (for example, one taken after electrode drift) distorts the entire trend; long-term monitoring requires periodically refreshing the reference frame;
• Not applicable to static objects: when the object itself does not change and no suitable reference state exists (as in one-shot structural inspection), relative imaging has nothing to work with.

Absolute Imaging: The Distribution Is the Object

Basic Principle

Absolute imaging inverts the absolute conductivity distribution at one instant from a single data frame. Since the problem is nonlinear, iterative methods are typically used:

Initial guess σ₀ → forward solve F(σₖ)
→ compare residuals F(σₖ) − V
→ update along the sensitivity direction σₖ₊₁ (Gauss–Newton, NOSER, etc.)
→ iterate to convergence with Tikhonov / TV / Laplace regularization

For 2D problems, the D-bar framework (supported by
uniqueness theory) is also available and is widely
used in research on lung imaging, among others.

Why Absolute Imaging Is Harder

1. Exacting requirements on forward-model accuracy. The reconstruction directly depends on how accurately the real physical system has been modeled:

Factors that must be modeled accurately:
├─ Sensor geometry (vessel diameter, electrode width, axial positions)
├─ Electrode contact impedance z (often unknown and time-varying)
├─ Boundary conditions (vessel wall, insulation, external conductors)
└─ Gain/phase calibration of the measurement system

Any deviation in these terms shows up in the image as artifacts or positional bias — errors that difference imaging cancels out all enter the picture in absolute imaging.

2. More severe ill-posedness. Absolute inversion is highly sensitive to errors in boundary data; central resolution is markedly lower than near-wall resolution, and depth-direction localization error is hard to avoid.

3. Contact impedance is the weak point of long-term operation. Electrode polarization, corrosion, and fouling all cause contact impedance to drift — and it is precisely the parameter hardest to know online. This is the fundamental reason long-term online absolute imaging requires periodic calibration.

Mitigation Strategies

• Offline calibration: calibrate the system against standard solutions of known conductivity to estimate contact impedance and gain;
• Structural priors / grey-box methods: incorporate known vessel walls, internals, or approximate target shapes as constraints, substantially compressing the solution space;
• Multi-frequency / multi-modality fusion: exploit differences of the target across frequencies and physical fields to improve the stability of absolute inversion.

Relative vs. Absolute: Side-by-Side Comparison

Aspect	Relative (difference) imaging	Absolute imaging
Input data	Difference of two frames ΔV	Single frame V
Nature of inverse problem	Linearized approximation, better conditioned	Nonlinear, strongly ill-posed
Required model accuracy	Low (common-mode errors cancel)	High (geometry / contact impedance / calibration must be right)
Image content	Conductivity change distribution	Absolute conductivity distribution
Typical artifacts	Reference-state leakage	Modeling-error artifacts, positional bias, central blurring
Online real-time capability	High (single linear step)	Low (iterative); optimization needed for online use
Reference state	Required	Not required
Typical applications	Pipeline multiphase-flow monitoring, filter/reactor process tracking, observation of changing processes	One-shot concrete/geological assessment, static objects without reference, absolute conductivity determination

In one sentence: relative imaging trades absolute values for robustness and real-time capability; absolute imaging trades demanding model requirements and computation for reference-free operation and absolute values.

Quantification: From Images to Numbers

Many industrial applications care about numbers, not just pictures: phase fraction, concentration, water content. The two imaging routes play different roles on this path:

• Relative imaging → captures the evolution of change (bubble formation and collapse, plume migration, bed expansion and contraction). With calibration, integrating the image sequence can also estimate the change in phase fraction corresponding to the observed changes;
• Absolute imaging → yields an absolute conductivity field, which mixing rules (e.g., Maxwell, Bruggeman equations) can convert into absolute estimates of phase fraction / porosity / concentration — something relative imaging cannot provide;
• The mainstream engineering practice: absolute imaging calibrates the baseline + relative imaging tracks the change — the absolute calibration sets the “zero mark of the scale,” high-frame-rate differencing traces the “process curve,” and together they form a complete measurement with both an absolute anchor and process detail.

Practical Selection Guide

Your scenario	Recommended route
Online gas–liquid / gas–solid two-phase flow monitoring in pipelines	Relative imaging (the default)
Process tracking in filters, reactors, and packed beds	Relative imaging + periodic reference-frame refresh
One-shot static assessment such as concrete compaction or dam seepage	Absolute imaging + structural priors
Absolute phase-fraction / concentration values required	Absolute imaging + mixing-rule conversion, with standard-solution calibration
Long-term online monitoring over months	Relative imaging as the mainstay; watch electrode drift and recalibrate periodically

Two field observations to add:

1. A reference frame is not “set and forget.” Temperature, electrode condition, and medium background all drift slowly; long-running projects should put “reference-frame refresh / recalibration” into the maintenance plan;
2. First be clear about what question you need answered. If you only care about “the pattern of change,” do not take on the complexity of absolute imaging; if you need “absolute numbers,” budget for calibration and model verification from the start — choosing the wrong route is a common cause of project rework.

Next Steps

Read Soft Field and Ill-Posedness in ET to understand the root of the inverse problem’s difficulty;
View ERT Technology Guide for an overview of industrial applications;
View Phase Fraction Measurement by Tomography for quantification and conversion methods;
Or contact us directly with your application scenario — our engineers will help you choose between “absolute values” and “process tracking” and define the right imaging route.

---

Core Perspective of This Article:

Relative and absolute imaging are not a matter of better or worse — they are two different problem definitions. Relative imaging differences away systematic errors, trading absolute values for robust real-time process monitoring; absolute imaging confronts the full inversion, trading calibration and priors for absolute values. Understanding the boundary between the two is what makes a sound technology choice between “process” and “numbers.”

The eTomography Pro software from Tianjin Youyi integrates multiple classic and intelligent reconstruction algorithms that can be switched and compared on the same dataset with one click, making it easy to evaluate different imaging strategies on real projects. Its companion ERT data acquisition systems support raw data recording and offline replay, providing the data foundation for reference-frame management and offline algorithm reproduction.