Since image compression formats like JPEG2000 are included in your evaluation, am I right in the assumption that in your experiments, you compressed each file individually, and finally summed the sizes of the compressed files?
I wonder whether there is a large difference between compressing files individually and compressing all files together (like tar'ed together).
In the latter case, I expect that LZMA or some window-less approach should do quite a good job.
Also, did you create a zstandard compression dictionary tailored for your use case to optimize the compression of zstandard?

Also I wonder which formats support to what extent the decoding/visualizing of regions of the image without decompressing the whole image.