Add artifact removal capability

.gitignore

@@ -15,7 +15,7 @@
 pyproject.toml.bak
 _version.py
 .ruff_cache/
-.pytest_cache/
+.pytest_cache
 htmlcov/
 
 ### Linux ###

.pre-commit-config.yaml

@@ -4,7 +4,7 @@ repos:
     rev: v5.0.0
     hooks:
       - id: check-added-large-files
-        args: [--maxkb=8192]
+        args: [--maxkb=30000]
       - id: check-json
       - id: check-merge-conflict
       - id: check-toml

README.md

@@ -1,37 +1,33 @@
-[![Documentation Status](https://readthedocs.org/projects/cylindricalgeometrycorrection/badge/?version=latest)](https://cylindricalgeometrycorrection.readthedocs.io/en/latest/?badge=latest)
-
 # neutron-geomcorr (Neutron Geometry Correction)
 
+[![Documentation Status](https://readthedocs.org/projects/cylindricalgeometrycorrection/badge/?version=latest)](https://cylindricalgeometrycorrection.readthedocs.io/en/latest/?badge=latest)
+
 Formerly known as CylindricalGeometryCorrection, this library performs geometric corrections for neutron transmission imaging of cylindrical samples (both solid and hollow).
 
-# Principle
-Any homegenous cylindrical sample placed in a beam (neutron for example) will show a much higher transmission signal
-near the edge seen by the beam, compare to the center. This is simply due to the fact that the beam has to go through
-more material at the center compare to the side.
+## Principle
+
+Any **homogeneous** cylindrical sample placed in a beam (neutron for example) will show a much higher transmission signal near the edge seen by the beam, compare to the center. This is simply due to the fact that the beam has to go through more material at the center compare to the side.
 
-ATTENTION: This library only works with cylinder placed in the vertical position!
+> ATTENTION: This library only works with **cylinder** placed in the **vertical** position!
 
 The following figure illustrate this.
 
 ![image](documentation/source/_static/homogeneous_cylinder_2d_view.png)
 
-In order to correctly analyze data for those samples, one must cancel this cylindrical effect by "making" the sample
-flat related to the direction of the beam.
-
-The user needs to specify the position of the center as well as the radius of the cylinder. The program will then produce
-an image corresponding to the same rectangular sample.
-
+In order to correctly analyze data for those samples, one must cancel this cylindrical effect by "making" the sample flat related to the direction of the beam.
+The user needs to specify the position of the center as well as the radius of the cylinder.
+The program will then produce an image corresponding to the same rectangular sample.
 Such samples are called homogeneous because they are made of only one uniform and homogeneous material.
-
 But program also work with inhomogeneous sample for which the cylinder is hollow such as shown here.
 
 ![image](documentation/source/_static/inhomogeneous_cylinder_2d_view.png)
 
 Program works the same way, user needs to specify center and inner and outer radius of material sample.
 
-# Installation
+## Installation
+
+### Using Pixi (Recommended for Development)
 
-## Using Pixi (Recommended for Development)
 ```bash
 # Install pixi if you haven't already
 curl -fsSL https://pixi.sh/install.sh | bash
@@ -47,17 +43,23 @@ pixi install
 pixi run python
 ```
 
-## Using pip
+### Using pip
+
 ```bash
 pip install neutron-geomcorr
 ```
 
-## Using conda
+> After package is published to PyPI.
+
+### Using conda
+
 ```bash
 conda install -c neutronimaging neutron-geomcorr
 ```
 
-# Usage
+> After package is published to conda.
+
+## Usage
 
 ```python
 from neutron_geomcorr.geometry_correction import GeometryCorrection
@@ -76,19 +78,22 @@ o_cgc.correct()
 corrected_images = o_cgc.list_data_corrected
 ```
 
-# Development
+## Development
+
+### Running Tests
 
-## Running Tests
 ```bash
 pixi run test
 ```
 
-## Building Documentation
+### Building Documentation
+
 ```bash
 pixi run build-docs
 ```
 
-## Building Package
+### Building Package
+
 ```bash
 # For PyPI
 pixi run pypi-build
@@ -97,5 +102,6 @@ pixi run pypi-build
 pixi run conda-build
 ```
 
-# Tutorial
+## Tutorial
+
 It's recommended to follow the notebook tutorials in the `notebooks/` directory to learn how to use the library

documentation/derivation.md

@@ -0,0 +1,128 @@
+# Cylindrical Geometry Correction for Transmission Data
+
+This document presents the methodology for applying cylindrical geometry correction to transmission data, specifically in the context of neutron or X-ray imaging where Beer-Lambert law governs the attenuation of intensity through a material.
+
+---
+
+## Background: Beer-Lambert Law for Transmission
+
+The transmitted intensity $ I(x, y) $ at a position $(x, y)$ through a material is described by the Beer-Lambert law:
+
+$$
+I(x, y) = I_0 \cdot e^{-\mu \cdot L(x, y)}
+$$
+
+where:
+
+- $ I(x, y) $ is the transmitted intensity at position $(x, y)$,
+- $ I_0 $ is the incident (background) intensity,
+- $ \mu $ is the linear attenuation coefficient (accounting for both absorption and scattering),
+- $ L(x, y) $ is the chord length through the material at position $(x, y)$.
+
+For a vertical cylindrical sample, the system can be simplified by considering only the horizontal coordinate $ x $, assuming uniformity along $ y $:
+
+$$
+I(x) = I_0 \cdot e^{-\mu \cdot L(x)}
+$$
+
+where $ L(x) $ is the chord length through the cylinder at position $ x $.
+
+---
+
+### Transmission and Attenuation Coefficient
+
+The normalized transmission $ T(x) $, also known as the normalized radiograph, is defined as the ratio of transmitted intensity to incident intensity:
+
+$$
+T(x) = \frac{I(x)}{I_0} = e^{-\mu \cdot L(x)}
+$$
+
+From this relation, the attenuation coefficient $ \mu $ can be calculated as:
+
+$$
+\mu = - \frac{\ln(T(x))}{L(x)}
+$$
+
+---
+
+## Practical Considerations
+
+In practice, ideal background measurements $ I_0 $ may not be available or may be imperfect, resulting in transmission values $ T(x) > 1 $, which leads to negative or non-physical values for $ \mu $. To address this, two robust methods for estimating $ \mu $ are presented below.
+
+---
+
+#### Method 1: Discrete Method (Two-Column Approach)
+
+This method leverages transmission values at two distinct positions $ x_1 $ and $ x_2 $ on the cylinder to estimate a physically meaningful attenuation coefficient.
+
+Given:
+
+$$
+T(x_1) = e^{-\mu \cdot L(x_1)} \quad \text{and} \quad T(x_2) = e^{-\mu \cdot L(x_2)}
+$$
+
+Taking the ratio:
+
+$$
+\frac{T(x_1)}{T(x_2)} = \frac{e^{-\mu \cdot L(x_1)}}{e^{-\mu \cdot L(x_2)}} = e^{-\mu \cdot (L(x_1) - L(x_2))}
+$$
+
+Solving for $ \mu $:
+
+$$
+\mu = -\frac{1}{L(x_2) - L(x_1)} \cdot \ln\left(\frac{T(x_1)}{T(x_2)}\right)
+$$
+
+**Interpretation:** This approach ensures that the calculated attenuation coefficient $ \mu $ is always positive and physically meaningful, as it relies on relative transmission differences rather than absolute values. The choice of $ x_1 $ and $ x_2 $ should correspond to positions with reliable transmission data.
+
+---
+
+#### Method 2: Iterative Method (Global Regression with Nuisance Term)
+
+An alternative approach involves performing a global regression across all spatial positions $ x $ to estimate $ \mu $ while accounting for nuisance parameters (such as background fluctuations or instrumental effects). This iterative method fits the model:
+
+$$
+T(x) \approx e^{-\mu \cdot L(x)} \cdot N(x)
+$$
+
+where $ N(x) $ represents a nuisance term that captures deviations from ideal background normalization.
+
+The regression optimizes the parameters $ \mu $ and $ N(x) $ to minimize the discrepancy between measured and modeled transmissions across the entire dataset. This method is particularly useful when background measurements are noisy or non-uniform.
+
+---
+
+### Correction Application per Wavelength (TOF Bin)
+
+It is important to note that the attenuation coefficient $ \mu $ and the corresponding correction factor are wavelength-dependent (i.e., vary with Time-of-Flight (TOF) bins). Therefore, the correction must be applied separately for each wavelength bin to accurately account for energy-dependent attenuation.
+
+---
+
+## Cylindrical Sample Thickness Correction
+
+Once $ \mu $ is determined, the goal is to transform the measured transmission $ T(x) $ corresponding to the chord length $ L(x) $ at position $ x $ to a uniform thickness corresponding to the cylinder diameter $ D $.
+
+Given:
+
+$$
+T(x) = e^{-\mu \cdot L(x)}
+$$
+
+We define the corrected transmission for uniform thickness $ D $ as:
+
+$$
+T^D(x) = e^{-\mu \cdot D}
+$$
+
+The correction factor $ C(x) $ to be applied to the measured transmission is:
+
+$$
+C(x) = \frac{T^D(x)}{T(x)} = e^{-\mu \cdot (D - L(x))}
+$$
+
+This factor adjusts the transmission data as if the sample thickness were uniform and equal to the diameter $ D $.
+
+> **Note:** Since the attenuation coefficient $ \mu $ is derived using differences in chord lengths, the correction factor $ C(x) $ is dimensionless and physically consistent.
+
+---
+
+This methodology provides a robust framework for correcting cylindrical sample transmission data, enabling accurate attenuation coefficient estimation and uniform thickness correction across all wavelengths.