Sony FCB camera blockWhen a Sony visible-light zoom camera exhibits white balance color cast, in the vast majority of cases it is not a hardware problem. It is more like a well-intentioned assistant using the wrong method: the AWB algorithm is constantly “guessing” the color temperature of the current light source based on frame statistics, and then applying color correction according to that guess. When the frame happens to match the algorithm’s statistical assumptions, the guess is accurate and colors look natural; when the frame is misaligned with those assumptions, the algorithm confidently makes an incorrect correction, and color cast arises.
The key to understanding AWB color cast in Sony visible-light zoom cameras is not to memorize which types of scenes are “prone to cast,” but to understand: what assumption stands behind each generation of AWB algorithm, and under what conditions that assumption collapses.
I. Four Generations of AWB, Four Increasingly “Tolerant” Assumptions
AWB technology has evolved through four generations since the early 2000s. Each generation is essentially using more complex statistical models to approximate the same problem: without a known white sheet, how can the light source color be inferred from the image itself?
The first-generation gray-world algorithm had the simplest assumption.
In a sufficiently colorful image, the averages of the R, G, and B channels should approach the same neutral gray value. It does not need to identify “white”; it only needs to perform one global average calculation and then pull the three channel gains equal. It is fast and simple to implement, but its Achilles’ heel is equally simple: no single color in the frame can be large enough to pull the average away from neutral gray.
The second-generation white-point detection algorithm took a different approach.
The brightest pixel in the frame is, ideally, the reflection of a white object under the light source. Find it, and you find the light source color. The fragility of this assumption is that “brightest” does not equal “whitest”—metallic reflections, bulbs, and LED point sources may all be brighter than a true white object, and they carry color.
The third-generation multi-target algorithm no longer relies on a single assumption. Instead, it maintains multiple detection mechanisms simultaneously and arbitrates among cues such as gray world, white point, and gray edges, attempting to converge to a reasonable result under more complex lighting.
The fourth-generation deep learning and AI algorithms use neural networks to learn from large amounts of data “what kind of frame should correspond to what color temperature,” further improving accuracy in complex environments at the cost of requiring training data and compute.
The progress of the four generations is essentially an improvement in the robustness of assumptions. However, the mainstream strategy in the built-in ISP of Sony visible-light zoom cameras is not a pure fourth-generation end-to-end model, but a hybrid architecture of gray world + white-point detection + multi-target algorithm, with third-generation multi-target arbitration as the backbone. This means: when both underlying cues—gray world and white-point detection—are simultaneously misled by the scene, multi-target arbitration loses a reliable basis for arbitration, and color cast occurs.
II. Four “Assumption Collapse” Scenarios for Color Cast in Sony Visible-Light zoom cameras
Scenario 1: Large-area single color—gray world gets “pulled off.”
The gray-world algorithm implies a premise: the scene’s color distribution must be balanced enough for the average to represent neutral gray. When the frame is dominated by a single high-saturation color, the R, G, and B averages no longer point to neutral gray but tilt toward the dominant color. The algorithm misreads the tilted average as “the light source caused a color cast,” and therefore applies reverse compensation; as a result, the image is pushed toward the complementary color of the dominant color.
Patent literature describes this clearly: the gray-world method cannot correctly estimate light source color in images containing large colored areas or objects, and the estimate will be “over-biased toward the color of that region or object.” In actual shooting scenarios with Sony visible-light zoom cameras: when green plants or grass fill the frame, the algorithm judges “green cast,” and after compensation the image becomes magenta; when blue sky dominates, after compensation it becomes yellow; when a red brick wall or yellow wall dominates, it shifts toward cyan; large bright white areas in snow scenes may instead be pushed toward blue under gray world. The green solder mask on PCBs in industrial AOI inspection and green surgical drapes in medical field cameras belong to the same type of collapse.
Scenario 2: Mixed light sources—the “dilemma” of color temperature estimation.
Indoor tungsten lamps coexist with daylight through the window, or fluorescent and incandescent lamps are mixed; different regions of the frame inherently have inconsistent color temperatures. Gray world and white-point detection can only provide one global light source estimate. Facing two real light sources, the algorithm is forced to compromise between two color temperatures, or to switch back and forth between frames, causing the overall image to be yellow or blue, or even causing color “jumping” as the lens moves.
The color cast here is not “calculating a certain color temperature wrong,” but rather that the algorithm architecture does not support spatial separation of multiple light sources. A single set of gains cannot simultaneously correct two color-temperature regions; any compromise leaves at least half the frame wrong.
Scenario 3: No reliable white point—white-point detection “calls a stag a horse.”
White-point detection requires a neutral white or gray surface illuminated by the light source to exist in the frame. When there is no white object at all in the scene (observation against an all-black background, solid-color product inspection), or when the brightest point itself is a colored highlight (metallic reflection, sodium lamp shade, LED indicator), white-point detection treats that colored highlight as “white” to calibrate the light source. When a metallic reflection is greenish, the algorithm thinks the light source is green and compensates toward magenta, and the entire frame becomes magenta. Under low illumination, noise further interferes with white-point judgment; AWB slowly drifts, and colors swing back and forth like “breathing.”
Scenario 4: Special spectra and non-convergence—blind spots outside the algorithm’s curves.
The spectra of sodium lamps and low-pressure mercury lamps differ greatly from daylight and blackbody radiation. The color temperature curves built into AWB cannot match them with conventional models, and color cast can be very severe. Sony official documentation explicitly notes: when mercury or sodium lamps are used as light sources, accurate white balance cannot be obtained due to the characteristics of the light source itself; custom white balance or a dedicated mode is recommended. The FCB-ER9500 technical manual does list Sodium Lamp Mode (Fix / Auto / Outdoor Auto), an exit specifically prepared for such “out-of-curve light sources.”
Another situation easily mistaken for a malfunction is AWB that has not yet converged: on first startup after power-on, or when a large zoom causes dramatic changes in frame content, the algorithm is in its learning phase, and brief color cast may appear and recover on its own after stabilization. This is not an algorithm judgment error, but judgment that has not yet been completed.
III. The Logic of Countermeasures: Switch from “Letting AWB Guess” to “Telling AWB the Answer”
Once the root cause of color cast is understood, the solution is no longer “repeatedly tuning AWB parameters and hoping for luck,” but choosing a white balance path that does not rely on the broken assumptions, according to the scene’s statistical characteristics.
When there is a reliable neutral reference in the scene, One Push WB is the most direct solution. Point at a white sheet or gray card and trigger once; the algorithm uses this real white point as an anchor to calculate gains, bypassing the guessing steps of gray world and white-point detection. Sony official recommendations also recommend custom white balance in mixed-light scenes, because it can faithfully reproduce white under multiple light sources.
When the light source type is known and stable, preset modes are more reliable than AWB. The FCB-ER9500 provides fixed presets such as Indoor, Outdoor, and Color Temp (2300K–15000K); indoor 3200K and outdoor 5800K are locked directly, no longer letting the algorithm re-estimate every frame. The cost is sacrificing automatic adaptation to light source changes, in exchange for determinism.
When the light source color temperature drifts but the scene itself has no large-area dominant color, ATW is a more suitable choice than standard AWB. ATW continuously tracks color temperature changes, and ATW WIDE covers a wider color temperature variation range, making it suitable for outdoor scenes where light changes slowly over time. But if a large-area single color exists in the frame, ATW will also be pulled off by the dominant color—mode selection cannot replace judgment about the scene’s statistical characteristics.
When the light source is a sodium or mercury lamp, the dedicated mode is the only correct entry point. This is not a problem that “parameter tuning can solve”; conventional color temperature curves themselves do not cover such spectra. You must switch to Sodium Lamp Mode and let the algorithm use correction logic customized for that spectrum.
In one sentence, the logic of countermeasures is: AWB is suitable for scenes with “balanced color distribution, a neutral reference, and a single light source”; once any one of these three conditions is broken, you should proactively switch to One Push, a preset, or a dedicated mode, and take judgment back from the algorithm.
Conclusion
AWB color cast in Sony visible-light zoom cameras is ultimately a footnote to an engineering commonplace: the reliability of an automatic algorithm never exceeds the applicability of the statistical assumptions it relies on. Gray world assumes “the world averages to gray,” white-point detection assumes “the brightest is white,” and multi-target arbitration assumes “at least one cue is trustworthy.” When the shooting scene dismantles these assumptions one by one, color cast is not a defect of the module, but an inference the algorithm makes under insufficient information—logically self-consistent but factually wrong.
When color cast occurs, first check whether there is a large-area single color in the frame, whether there is a reliable white reference, and whether the light source is single and conventional. After answering these three questions, the answer to whether to use One Push, a preset, or Sodium Lamp Mode is already there.
FAQ
Q: Since AWB corrects automatically, why are manual white balance or One Push more recommended in industrial inspection and medical scenarios?
A: Because these two types of scenarios systematically undermine the statistical premises on which AWB depends. The green solder mask in industrial AOI and the green surgical drapes in medical fields are large-area single colors, and gray world will be pulled toward the complementary color; solid-color product inspection or observation against an all-black background often lacks a reliable white point, so white-point detection loses its anchor. In such scenes, AWB’s “automatic” operation is not improving accuracy but continuously introducing an unpredictable variable. One Push locks gains with one real white-point calibration, and manual presets directly fix the color temperature. What is sacrificed is adaptability to light source changes; what is gained is a repeatable, verifiable color reference—for applications that need color judgment, determinism matters more than being “smart.”
Q: What is the difference between the FCB-ER9500’s Sodium Lamp Mode and ATW? When should each be used?
They solve problems at different levels. ATW is color temperature tracking: it still works on conventional color temperature curves, but continuously follows changes in light source color temperature. It is suitable for outdoor scenes where daylight drifts slowly over time, and ATW WIDE covers a wider color temperature range. Sodium Lamp Mode is spectral adaptation: the spectrum of sodium lamps differs too greatly from conventional light sources, and conventional color temperature curves simply cannot match it; Sodium Lamp Mode uses correction logic specifically defined for that type of spectrum. The criterion is straightforward: if the light source is daylight, tungsten, or conventional fluorescent and the color temperature changes, use ATW; if the light source itself is a sodium or mercury lamp, ATW cannot solve the spectral mismatch, and you must switch to Sodium Lamp Mode.
References
1. US Patent US20240196104A1, *Brightness based chromaticity weighting for improved illuminant color estimation for auto white balancing*
2. Sony, *FCB-ER9500 Specifications* (white balance modes: Auto, ATW, Indoor, Outdoor, One Push WB, Manual WB, Outdoor Auto, Sodium Lamp Mode Fix/Auto/Outdoor Auto)
Author
Danny Wong, Co-founder & Chief Technology Officer of Ruilong Electronic, compiled in September 2026.
Shenzhen Ruilong Electronic Information Technology Co.,Ltd., an officially authorized distributor of Sony (China), has 17 years of experience in HD video secondary development and implementation. It has independently developed IP / SDI / HDMI / USB / MIPI multi-type interface encoding control boards, and also provides a DSLR camera network control board compatible with the ILX-LR1. It can provide pre-sales solution consulting, on-site debugging, and after-sales technical support, support connection with Sony's original factory technical resources, and solve project problems on site.