another salvaged thread from DPReview:
Hot pixels are damaged sensor pixels which result in very bright image pixels that do not correspond to imaged objects. Such damage is known to happen by cosmic rays. It cannot be avoided in practice. The sensor may have tens of such hot pixels, which usually go unnoticed in most photographic work.
The HPS schemes try as
Step 1) to identify such false data in the image from a hot pixel on the fly. And in
Step 2) substitute a non-objectionable value for the hot pixel to repair the damaged image.
HPS schemes have been criticised for being to aggressive in removing valid image data and because the substitute data implied image damage: for example, rather than false stars from the hot pixels, the corrected image had missing stars.
Analyzing black frame images confirms that in the D850 HPS is active for exposure times greater or equal to 1/4s . The HPS scheme is also active in the D850 for ISO 6400 and up. It cannot be turned off by the user. "Long exposure NR" is off for the analysis below.
Analyzing the black frame data in detail, a very clever hot pixel recognition criterion emerges for Step 1) in the D850:
D_threshold = parameter1 * D_max_24 + parameter2
I define the net digital pixel value D_ = DN - D_black with the read out raw digital raw number DN before correction by the black level D_black
If the actual D_ of the center pixel exceed the D_threshold value, it is declared as hot pixel. D_max_24 is the maximum occurring on any of the other 24 other pixels in the 5x5 patch surrounding the pixel being examined. Analysis for D850 black frame shows without exception that parameter1 =2 and parameter2=400, (and D_black=400).
The value of parameter1 makes a lot of sense. It implies that a lens producing such an unreasonably sharp image would exceed MTF50 > 100lp/mm. No such lens is expected to be in use for D850. None may ever be available for covering FX .
The value of parameter2 also makes a lot of sense. It implies that the read noise should not trigger a false hot pixel identification. A similar argument goes for D_black. The value of parameter2 can be gleaned from the black frame data at higher ISO where false hits appear. Nikon engineers chose to turn the HPS on for higher ISO values than this: from ISO 6400 HPS is always on for the D850.
Because of the particular parameter values, it is very possible that the threshold criterion is actually implemented as
DN_threshold = 2 * DN_max_24
I have not yet gleaned the precise scheme for Step 2) . The substituted value at the hot pixel appears always to fall between DN_max_24 and 2 * DN_max_24 . This is a moderately bright pixel value, in reasonable relation with neighbor pixel values. It is a very acceptable guess of the noise or image brightness that should have appeared at the hot pixel site with a real lens.
graph 1: HPS is ON. A few pixels have a value just 1 below the threshold values for HPS . (it is on also for ISO greater than 5080)
graph 2: HPS is OFF till 1/5 s exposure and of up to ISO 5080. Without HPS there are lots of pixel values above the HPS threshold: plotted bold.