horizontal mosaicking of two different orbits

This is some sort of "proof-of-concept" for the mosaic problem we have. I didn't have any suitable S1 images ready, so I tested it on S2, and it seems to work.

Some notes:

• For identifying the overlap area, I only rely on the out-of-orbit (=nodata) value; no spatial information from the footprints/databases etc. is used. If this specified value occurs somewhere else in the image, results will probably be weird.

• Images have to have the same projection, the same height and overlap exactly in north-south direction. East-West can be different.

• It is possible to apply some sort of "buffer" to the overlap area to remove unwanted border pixels.

• It can be only used for two input images.

• It can only be used to blend images together horizontally.

Usage

This script can be used from the command line or imported into python scripts.

usage: horizontal_blending.py [-h] [-ooo_A VALUE] [-ooo_B VALUE]
                            [-o OUTPUT_FILE] [-b BUFFER_SIZE_PIXEL]
                            image_A image_B

horizontal seamless mosaicking of two images

positional arguments:
image_A               Image A
image_B               Image B

optional arguments:
-h, --help            show this help message and exit
-ooo_A VALUE, --out_of_orbit_A VALUE
                        Out-of-orbit value for Image A
-ooo_B VALUE, --out_of_orbit_B VALUE
                        Out-of-orbit value for Image B
-o OUTPUT_FILE, --output_file OUTPUT_FILE
                        /path/to/output_file
-b BUFFER_SIZE_PIXEL, --buffer_size_pixel BUFFER_SIZE_PIXEL
                        size of border to remove (in pixel)

or

from horizontal_blending import horizontal_blending

horizontal_blending(image_A: str,
                    image_B: str,
                    output_file: str = None,
                    nodata_A: int = 0,
                    nodata_B: int = 0,
                    buffer_size_pixel: int = None) -> str:

Examples

Image 1

Image 2

merged without blending

merged with blending

Sigmoid functions

For calculating the overlap, I used the sigmoid function:

Parameters were picked with trial and error, see this notebook

Blending function:

alpha = sigmoid(len_overlap)
out = row_A * (1.0 - alpha) + row_B * alpha

Dependencies

A lot unfortunately, since I quickly threw this together:

• rasterio (tested with 1.0.23)
• numpy (tested with 1.16.4)
• openCV (tested with 4.1.0.25)
• GDAL command line utilities available on PATH (tested with 2.2.3)