Image brightness and luminance Lab — exposure, gamma and clipping

Try a comparison

Run Identity and save it as A. Compare Linear exposure +1 EV with Encoded gain 2×. Tap an image or enter x/y to inspect C and Y at the same coordinate before and after processing. Switch to gradients to inspect changes at boundaries. Shared URLs save input conditions, not result A or the inspection coordinate.

Controls are available in Japanese and English. No local Python installation is required.

Pixel values and relative luminance

C is an encoded grey value from 0 to 1. This Lab assumes neutral sRGB, R=G=B, and decodes it to linear relative luminance Y. C=0.5 gives Y≈0.214. These values are not calibrated camera or display measurements and cannot be reported as physical luminance in cd/m².

Processing order

EV multiplies linear Y by 2^EV. Clamp to 0–1 and encode to sRGB; then apply encoded gain and offset, contrast about 0.5, another clamp, and gamma. γ>1 lifts intermediate values. There is no clamp immediately after gain and offset. Darkening later cannot recover distinctions lost in the first clip.

Y = decode_sRGB(C) × 2^EV
C₁ = encode_sRGB(clamp(Y, 0, 1))
z = gain × C₁ + offset
z = 0.5 + contrast × (z − 0.5)
C_out = clamp(z, 0, 1)^(1/gamma)

Read the measurements

Histograms count actual pixels in 32 equal bins, including the left edge and including 1 only in the last bin. C and Y are counted separately. Mean Y averages decoded pixels. Gradients use right and down forward differences over 159×119 positions. Images use fixed C=0–1 and gradients fixed 0–√2 scales. Clipping counts strict values below 0 or above 1 before a clamp, distinct from final endpoint occupancy. The unique count removes overlaps between stages.

Diagram 1 · Use the button to switch views

Connect to computer vision

Compare the low-contrast and low-texture presets. The same synthetic scenes appear in extraction, matching and optical-flow Labs. Gradient changes alter detector inputs; illumination changes can violate brightness constancy in flow. A histogram alone does not measure detection, correspondence or tracking accuracy.

Reproduce and explore

• Feature extraction Lab
• Feature matching Lab
• Optical flow Lab
• Source package
• Computed results JSON
• Model definition
• sRGB relative luminance (W3C)