GPU Spectral Viewer
During the EU IST project CRISATEL the consortium made around one
hundred multispectral scans of paintings conserved in several
museums. The obtained high resolution images allow us to not only have
a nice and accurate color image but also to investigate in depth with
quantitative scientific measure the color structure of the work of
art.
One of these paintings in particular required special attention. It
was
Leonardo da Vinci Mona Lisa. In order to study it a
special
GPU oriented software was developed. This page
describes
GPU Spectral Viewer an innovative software for
multispectral image analysis specialy developped for
La
Joconde painting.
In order to understand the colorimetric computation used in GPU
Spectral Viewer you can read the multispectral section of
Couleur.Org.
We are going to present here all the colorimetric measures and
visualizations modules that are used in our application. This software
based on an
ad hoc developed 2D rendering engine which uses
OpenGL. This engine was chosen because it can run on several different
systems (Linux, Unix, MacOSX and various Windows) but needs a Shader
Models 3.0 compatible graphic card in order to work properly. A
specific C++ library based on a composition engine was designed for
this purpose. This library allows us to build applications from a new
set of widgets rendered using textures and frame buffer objects. The
result is a fast and portable GUI built with specific design which use
eye candy transparency and 3D effects.
This software includes the following features:
- All the processing are done by the GPU instead of the CPU. The
result is a real time computation even on large multispectral images
(2000x3000 13 planes images)
- It includes a new specific color management process.
- Color manipulation in RGB, XYZ and Lab color spaces
- It uses complex color distance like DE1994 and DE2000
- ...
In order to visualize all the videos presented here you need the
following video
codec: xvid
Multispectral information visualization
| | The multispectral information visualization module
is used in order to display the reflectance image as a set of 16
bits grey level images. Different LUT and a gamma correction factor
can be modified during the visualization process Video Low Res Video High Res |
Top Color Reconstruction
| | Based on
XYZ and Lab color spaces and on a specific color management
process, this module allows the user to interact in real time with a
virtual environment light. All the generated and displayed images
are HDR images (High Dynamic Range). Video Low Res Video High Res |
Top
Color Measure
| | We use this GPU
processing module in order to compute the resulting Lab color
image. The result is different from the simulation module because we
use the white reference as virtual illuminant to do the
computation. Using this new Lab image we can provide a more usable
and precise interface for color selection Video Low Res Video High Res |
Top
Sensor Simulation
| | The analysis
of a painting using color can not be precise using a simple color
reconstruction based on different virtual lights (metamerism
analysis for example). For this reason we provide another specific
module where we can choose interactively the sensitivity curves of a
virtual RGB sensor. These sensitivities can be based on existent
sensor or interactively inserted by the user. Video Low Res Video High Res |
Top
Color Segmentation 1
| | The segmentation
process investigated for this application is based on color
clustering and threshold approaches. It can be time consuming,
especially if we use complex color distance. Video Low Res Video High Res |
Top
Color Segmentation 2
Top
Color Segmentation 3
Top
Color Cloud Visualization
Top
Elevation
| | This module is very
useful for experts that want to analyse separately each color
components (luminance for example). Video Low Res Video High Res |
Top
