Nighttime sky/cloud image segmentation

With the spirit of reproducible research, this repository contains all the codes required to produce the results in the manuscript: S. Dev, F. M. Savoy, Y. H. Lee and S. Winkler, Nighttime sky/cloud image segmentation, Proc. IEEE International Conference on Image Processing (ICIP), 2017.

Please cite the above paper if you intend to use whole/part of the code. This code is only for academic and research purposes.

Code Organization

The codes are written in python and MATLAB.

Dataset

The nighttime image segmentation dataset can be downloaded from this link. A few sample images can be found in the folder ./images.

Core functionality

• color16Norm.m Generates the 16 color channels in the form of a MATLAB struct. All values are normalized.
• color16_struct.m Generates the 16 color channels in the form of a MATLAB struct.
• createSPImage.m Generates the quantised- and binary- image of our proposed method.
• createSPImageNumber.m Generates the quantised- and binary- image of our proposed method, based on the number of superpixels.
• gacal.m Implements the Gacal approach.
• global_th_novi.m Implements the Yang et al. 2009 approach
• internal_calibration.py Implements the internal calibration of our sky camera.
• local_th_novi.m Implements the Yang et al. 2010 approach.
• RGBPlane.m Extracts the red-, green-, blue- plane of an input image.
• score.m Calculates precision, recall, fscore and error of a binary output image.
• showasImageNovi.m Normalizes the image to a range [0,255].
• SPS_novi.m Implements the Liu et al. approach.
• undistort_WAHRSIS_imgs.py Undistorts our sky camera images; needed during the creation of the dataset.

Superpixel function

The various functions required in SLIC superpixel segmentation can be found in the folder ./SegmentationToolbox. The core functions of SLIC are re-distributed under GNU General Public License terms.

Reproducibility

In addition to all the related codes, we have also shared the generated results. These files are contained in the folder ./results.

Please run the following to generate the various figures and tables in the paper.

• Figure1.m Demonstration of the proposed segmentation algorithm.
• Figure3.m Computes the cloud coverage of the sample images of the dataset.
• Figure6.m Performance of the various color channels for nighttime image segmentation.
• Statistics of SWINSEG dataset.ipynb Computes the distribution of images in the image dataset.
• Table2.m Performance evaluation of various benchmarking algorithms.