This repository includes MATLAB codes and datasets used in our manuscript (A. Moghimi, T. Celik, A. Mohammadzadeh and H. Kusetogullari, "Comparison of Keypoint Detectors and Descriptors for Relative Radiometric Normalization of Bitemporal Remote Sensing Images," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 2021, doi: https://doi.org/10.1109/JSTARS.2021.3069919,) for relative radiometric normalization (RRN) of unregistered bi-temporal multi-spectral images.
The MATLAB code implements the fully automatic relative radiometric normalization methods for unregistered satellite image pairs based on the different feature detector-descriptors, presented in our manuscript.
For code and datasets, please take a look at the supplementary material.
The codes are developed and tested in MATLAB R2020a, with both OpenCV.3.4.1 and the VLFeat open-source libraries on a desktop computer with Intel(R) Core (TM) i7-3770 CPU @ 3.40 GHz, 12.00GB RAM, running the Windows 8.1. To use the codes, you need some prerequisites as follow:
- MATLAB 2020a
- OpenCV (3.4.1)
- VLFeat 0.9.21
you also need the required build tools for Windows, which is Visual Studio. Please see https://github.com/kyamagu/mexopencv for how to download and install OpenCV on your MATLAB software. Also, please see the https://www.vlfeat.org/ for downloading and installing VLFeat 0.9.21.
Getting Started After installing OpenCV 3.4.1 and VLFeat 0.9.21, it is enough to use only main.m for a quick start. Here are some examples.
In the path of MATLAB codes, please open main.m in your MATLAB editor press “run” or type “run Main” in the Command Window, and then press “Enter” as follows:
please select the subject and reference images from the opened windows as follows:
please select a detector for RRN from the opened menu as follows:
The process is terminated after its execution is naturally completed, and the final RRN results of RRN using the selected detector are demonstrated. Here are some visualized results
More configuration items for detector-descriptors can be found in functions SiftDetector.m (SIFT), SurfDetector.m (SURF), and feature_detector.m for (KAZE, AKAZE, ORB, SURF). The metrics (RMSE and NTG) are also provided in the score_index.m. Furthermore, RCS_Regression.m is a function that selects RCS from matches and RRN modeling. In addition, appendimages.m is used for visualition matches, and TIN-basedLocalAffine.m is a program of TIN-based local affine for blunder rejection from matches.
Keypoint-based RRN results on different datasets: (a) Subject (Sub.) and reference (Ref.) images in each dataset; (b) Normalized subject images (bottom) using different keypoint detectors/descriptors; and (c) The percentage of inliers, generated by the keypoint detectors/descriptors with different cross-correlation ranges.
-
Dataset 1 from Tabriz, Iran, shows typical characteristics of images captured in urban areas. link for download: https://1drv.ms/u/s!AvQPxeTMtP1Hai6X5Kd9IHrBz-4?e=QZEpXh
-
Dataset 2 from Cagliari, Italy, is mainly comprised of scenes with different land covers such as rural areas, mountains, vegetation (e.g., farmland and sparse forest), and water bodies and also shows heavy seasonal changes due to the vegetation transition and increase in the surface area of the water body. It is available on the GitHub repository as named (Dataset 2).
-
The bi-temporal images in Dataset 3 are from Daggett County, USA. The images cover mountainous regions with scattered vegetation patterns and a water reservoir (Flaming Gorge) and show temporal changes, largely due to cloud covers and their shadows. link for download: https://1drv.ms/u/s!AvQPxeTMtP1HbTnvKmGI3PD4g68?e=kPJhZm
-
Dataset 4 from Cape Town, South Africa, depicts the characteristics of images acquired on coastal areas. link for download: https://1drv.ms/u/s!AvQPxeTMtP1Hco5CN4YWbei5sz4?e=lQ4Hhw
-
Dataset 5 from Bamako, Mali, is comprised of a moderately high-resolution image pair covering a semi-urban area with diverse image contents. It is available on the GitHub repository as (Dataset 5).
Thanks to the EarthExplorer-USGS (https://earthexplorer.usgs.gov/) and Airbus Intelligence (https://www.intelligence-airbusds.com/en/8262-sample-imagery), I couldn't have finished my experiments without these carefully collected datasets. Much of this work referred to the open-source community (https://www.vlfeat.org/index.html) and (https://github.com/kyamagu/mexopencv), for which I'd like to thank all these authors. Also, I would like to thank Professor Turgay Celik (https://scholar.google.com/citations?user=FpJjjtIAAAAJ&hl=en), Professor Ali Mohammadzadeh (https://scholar.google.com/citations?user=C5bzZSsAAAAJ&hl=en), and Dr. Huseyin Kusetogullari (https://www.bth.se/staff/huseyin-kusetogullari-hku/) for their kind help and valuable advice.
If you find these codes and datasets helpful to you, please consider citing our papers.
-
A. Moghimi, T. Celik, A. Mohammadzadeh and H. Kusetogullari, "Comparison of Keypoint Detectors and Descriptors for Relative Radiometric Normalization of Bitemporal Remote Sensing Images," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, doi: https://doi.org/10.1109/JSTARS.2021.3069919.
-
A. Moghimi, A. Sarmadian, A. Mohammadzadeh, T. Celik, M. Amani and H. Kusetogullari, "Distortion Robust Relative Radiometric Normalization of Multitemporal and Multisensor Remote Sensing Images Using Image Features," in IEEE Transactions on Geoscience and Remote Sensing, doi: https://doi.org/10.1109/TGRS.2021.3063151.
Contributions and suggestions are highly welcome. Let's work together!