RAW format tolerance stretch limit: the signal-to-noise ratio collapse boundary when brightening shadows by +5EV and the mathematical mechanism of highlight recovery

1. The physical cost of shadow brightening +5EV: the riot of read noise and shot noise

When I pull the shadow slider to +100 in Lightroom, why is the image full of colorful noise and streaks?

1. Full well capacity (FWC) and signal-to-noise ratio cliff for weak signals:

• The photodiode under the microlens of each pixel can hold a limited number of electrons (usually 30,000~80,000 electrons);
• In the highlight area, the pixel collects tens of thousands of electrons, the signal is extremely strong, and the signal-to-noise ratio (SNR) is extremely high;
• In the extremely dark shadow area, the pixel only collected 10~20 photons during the exposure time;

2. Read Noise Floor:

• Every time the sensor reads data, the inherent thermal noise and conversion noise inside the chip are about 1~3 electrons;
• When you force-brighten the shadows by 5 stops in post (equivalent to amplifying the original signal 32 times):
• While the weak signal is amplified, the underlying fixed readout noise and photon shot noise are ruthlessly amplified 32 times simultaneously. The originally pure black and smooth area instantly collapses into Chroma Noise and horizontal banding!

2. The irreversibility of highlight overflow and the mathematical truth of the "highlight recovery" algorithm

It is often said in the photography industry that "digital preserves highlights, film preserves shadows":

1. Mathematical definition of digital saturation (Clipping):

• The numerical upper limit for 14-bit RAW files is 16383 (2^14 - 1);
• Once the amount of light entering exceeds the full well capacity of the pixel, electrons overflow, the reading recorded by the ADC is permanently locked at 16383, and all texture levels are completely lost to zero variance constant;

2. What exactly does “Highlight Recovery” in Lightroom do? ：

• Most of the time highlight bloom is "single channel bloom" (for example, the green channel is saturated and overflowed, but the red and blue channels are not full);
• The algorithm uses the data of unsaturated channels, uses the color space reverse color difference interpolation algorithm, and uses brain calculation to calculate the relative proportion of the overflow channel, and restores the faint gray-scale clouds in the sky;
• Absolute Limit: Once the three channels of red, green, and blue are all overflowed and saturated (all reaching 16383), no god-level algorithm can create details out of thin air, and the picture will only be left with ugly gray-white blocks!

3. Exposure to the right (ETTR) and tolerance guard golden workflow

1. Highlight zebra crossing is set to 100%+:

• Turn on the body zebra crossing, adjust the shutter and aperture so that the zebra crossing just flashes at the edge of the brightest reflective point to maximize the amount of physical light entering the sensor;

2. Resolutely shoot RAW format bracketing (Bracketing ±2EV) in large light ratio scenes:

• 32-bit HDR fusion is used in post-production to perfectly unify true details of highlights and high signal-to-noise ratio of shadows.

---