https://github.com/totti0223/deepstomata/
A three step image analysis program for quantification of stomatal aperture from bright field images.
- Identifying the coordinate of the stomata by HOG + SVM.
- Classifying the status (open, partially open, closed, false positive) by CNN.
- Pore quantification responsive to the object status.
- licence is not determined yet.
- planning to submit an article soon.
- contact me if you want any details. (Jan,27,2018)
Yosuke Toda, Ph.D (Agriculture)
Post Doctoral Researcher
Lab of Plant Physiology
Department of Science
Nagoya University, Japan
python>3
matplotlib==1.5.1
numpy==1.11.2
scipy==0.18.1
scikit_image==0.12.3
tensorflow==0.10.0rc0
Pillow==3.2.0
common==0.1.2
cv2==1.0
dlib==19.1.0
setuptools==32.3.1
-
Download this repository.
-
Unzip.
-
Open terminal.
-
Move to the Unzipped directory.
pip install .
-
Tensorflow must not be ver. 1.0.. Codes are not compatible.
-
Several packages such as cv2 and dlib cannot be installed via pip in anaconda environment. In such cases, comment out the requirements.txt like the following
#cv2 ==1.0
#dlib == 19.1.0
and install respectively via conda install
- In terminal
python
import bmicp
bmicp.cui("PATH/TO/THE/DIRECTORY_OR_IMAGES")
Analyze a directory containing 4 jpeg images in the example folder
import bmicp
bmicp.cui("PATH_TO_THE_EXAMPLE_FOLDER/examples")
Result overview in the terminal.
Image with stomata position.
Montage image of stomata candidate that is same size as input image.
Image with stomata position, class percentage.
Image with stomata position, class percentage, and segmented stomatal pore.
CSV files with quantified stomatal pores.
Statistics of classified object per image (no. of open, partially open, closed, false positive per image).
-
Migrating CNN code from tensorflow to keras
-
GUI
-
Registrating the package to PyPi (Packaging the CNN model exceeds the upload size limit of PyPI)