8.9 Dead pixels⧉
Defective photosites can create isolated dark or bright marks. A dead site has little or no useful response; a stuck site reports an abnormal fixed value, which need not be full scale. A hot site has unusually high dark current, most visible at long exposures and elevated temperature. ISO can amplify its recorded value but does not itself generate the leaking charge.
8.9.1 Where they come from⧉
Some defects originate in manufacture; others develop later. Dark current depends strongly on temperature and exposure duration, with the rate varying by sensor and defect. Avoid using one universal temperature-doubling rule to predict a particular camera.
8.9.2 Finding them⧉
Factory calibration can supply a defect map. DNG supports FixBadPixelsConstant and FixBadPixelsList opcodes, allowing correction to be described in the file. Calibration or a camera pixel-mapping function can update known defects. A matched dark frame helps identify abnormal dark response, but one bright sample is not proof of a defect. Subtracting a dark frame removes a repeatable offset while adding that frame's own random noise.
A camera that takes a matched dark exposure after a long photograph spends additional capture time measuring its own dark response. The exposure duration and camera settings should match, and temperature should be comparable. This explains the delay associated with that form of long-exposure noise reduction. A defect map and dark-frame subtraction address related but distinct problems: the map identifies unreliable sites, whereas subtraction estimates an additive pattern.
8.9.3 Fixing them⧉
Correction estimates the missing measurement. In a Bayer mosaic, use valid samples of the same CFA color; a nearby green is not a substitute for a missing red. For an isolated known defect, a median of nearby same-color samples is a simple robust estimate. Exclude other mapped defects, expand the neighborhood if needed, and treat clusters or rows separately. Correct early so demosaicking does not spread the error.
Blind outlier suppression can erase stars or small specular highlights along with hot pixels. Astrophotographers named the failure the star eater when, around 2016, a firmware change in some mirrorless cameras was widely reported to filter faint stars out of long exposures along with the hot pixels it targeted. A calibration map supplies evidence that a bright site is defective; brightness alone does not. Preserve the original raw measurements when possible so a later reconstruction can make a different choice.
A star and a hot pixel can both be brighter than all their immediate neighbors. An outlier detector based only on local brightness can remove either. A stored defect map provides evidence tied to sensor location; repeated captures can test whether the bright point moves with the scene or remains fixed on the sensor. Neither test is available in every situation. The correction should reflect the available evidence rather than treating all isolated bright samples as defects.