Implementation of the image analysis system described in Steven Bergner's thesis work on "Deformable structural models", also published at IEEE ICIP 2004.
Travis-CI status:
The system involves feature detection layers with convolutions and non-linear functions of the underlying color image data input.
A combination of stochastic search and local gradient descent with damping is used to optimize a graphical model of shape components that finds connections among maxima in the feature detection sensor layer and compares them to dynamically adjustable models of different ant species.
An overview of the main classes implemented in the src folder is given in
Chapter 5 of the thesis "Structural Deformable Models for Robust Object Recognition"
on page 61 (p. 81 in the PDF).
Also have a look at the auto-generated documentation.
Sensors can be convolutions or non-linear functions of source sensors in this bottom-up data flow architecture.
Shape part models of all species are related in stochastic scene graph.
Highly scoring paths indicate that the image data can be interpreted as a particular species.
A statistical model of relative shape position is attached to each edge in the stochastic scene graph.
A grid in (x, y, rotation, scale) feature space is used to determine local winners
Color samples of ant body and background are acquired interactively
Scatter plot of ant body color samples
Mahalanobis distance iso-surface of Gaussian model for background color samples
Posterior probability iso-surface of combined Gaussian model for ant body color samples
If you are lucky, you can install dependencies and just build
Linux:
sudo apt install freeglut3-dev mesa-common-dev libfox-1.6-dev libfftw3-dev libgl1-mesa-glx
OSX (using homebrew):
brew install fox fftw freeglut lua
Then:
./download-externals.sh
cd src
make -j4 release <OR> make -j4
This builds the deform binary in a architecture and build
variant specific subfolder, e.g. src/obj/LINUX, src/obj/LINUX_debug, or src/obj/DARWIN.
If that does not work, you may have to adjust build paths and compiler options in
src/Makebundle/<PLATFORM>.config.make or the remaining Makebundle infrastructure.
Libraries and external sources required to build and run the application:
- FOX toolkit (if building manually use at least
./configure --with-opengl) - FFTW3 - for fast Fourier transforms
- Eigen3 - for easy to use linear algebra
- System installation of OpenGL - to see the algorithm in action
- (optional) Camgraph - for EPS vector graphics export (deactivated)
Have a look at .travis.yml to see how dependencies are installed
automatically on the Travis-CI build-bot running Ubuntu 14.04 (precise).
Compilation under Windows 10 using its new Bash should also be possible (TODO update instructions here).
The algorithm works on image data, e.g., as collected by MCZ at Harvard Entomology.
Download of macro-photographic image data for personal use or research, or whichever type of usage permitted by the database website, can be done using a script (TODO) and wget.
Copyright 2003, 2017 Steven Bergner
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.