Starry starry Night, Foveon-wise

In the top layer of the Foveon F20 sensor made by DongBu Hi-tek in a far-off land, many if not all have a random scattering of over-sensitive sub-pixels (the "blue" ones). In many images, they are quite evident, especially in dark areas. Apparently, the Sigma proprietary converter also the on-board raw-JPEG converter know of these bad sub-pixels because they don't show up in RGB review images,

However, they become visible if the converter does not action the bad pixels list found in the raw meta-data, i.e. does not change the three values of the bad pixels to the average of the surrounding neighbor pixel values.

So here's an 878x641 px crop from an X3F (raw) image, converted by RawDigger

The "stars" are more easily seen in the expanded image.

In case you don't see them, here is the same image processed by a Threshold function which converts the image to pure black and white, the threshold being about 25% for this image.

Like many sensors, the Foveon has it's quirks which are secretly hidden by the manufacturer.

Not the sensor, but try viewing the raw image from a Pansonic m4/3 building shot and observe the lens distortion, not seen in the JPEG..

@xpatUSA has written:

In the top layer of the Foveon F20 sensor made by DongBu Hi-tek in a far-off land, many if not all have a random scattering of over-sensitive sub-pixels (the "blue" ones). In many images, they are quite evident, especially in dark areas. Apparently, the Sigma proprietary converter also the on-board raw-JPEG converter know of these bad sub-pixels because they don't show up in RGB review images,

However, they become visible if the converter does not action the bad pixels list found in the raw meta-data, i.e. does not change the three values of the bad pixels to the average of the surrounding neighbor pixel values.

So here's an 878x641 px crop from an X3F (raw) image, converted by RawDigger

The "stars" are more easily seen in the expanded image.

In case you don't see them, here is the same image processed by a Threshold function which converts the image to pure black and white, the threshold being about 25% for this image.

Like many sensors, the Foveon has it's quirks which are secretly hidden by the manufacturer.

Not the sensor, but try viewing the raw image from a Pansonic m4/3 building shot and observe the lens distortion, not seen in the JPEG..

This is pretty much the norm; most cameras simply replace known bad pixels with the median value of neighbors before writing the RAW. The Canon 10D at ISO 3200 "revealed" half of these pixels, because it did something that seems to be a banding noise correction that left clues. ISO 3200 is the same as ISO 1600 original digitization and gain, but doubles the raw numbers (throwing away a stop of headroom). So, all the raw values at ISO 3200 should be even numbers. However, Canon did something that looks like banding correction, and added or subtracted from values per row, either 01010101010101 etcetera, or 1111111111111 etc, so if you look at only the least significant bit, every row is either:

000000000000...
010101010101.... or
11111111111111

However, there were sparse variations:

000001000000...
010101000101.... or
1111101111011

... which can only be from applying a median replacement from a list of pixels, after the least significant bit alterations. Half of the mapped-out pixels were therefore displayed in the LSBs, and the other half are not visible because the medians are the same as the pattern.

I checked out the noise of your pixels, and it looks like it has no spatial correlation at all except that which is randomly due to chance. 3.34% of your pixels were bad, so I made an image the same size as a 256x256 crop from your image, and randomly made 3.34% of the pixels white, and they both had the same stats, and then the acid test: I binned 16x16, and the stats were still similar. It may look like there are large "black holes" in your image, but that is the normal look of purely random dots at that low of a density..