Veiling Glare

News: October 2021:  A new method using two exposures (without and with a special mask) provides an excellent measurement of veiling glare as well as two flavors of dynamic range. February 2018: Imatest presented a paper, “Measuring the impact of flare light on Dynamic Range,” at Electronic Imaging 2018. It can be downloaded here. Imatest 5.0+  The Uniformity and Uniformity Interactive modules include ISO 18844 flare measurements.

 

Related pages Stray light (Flare) – The small, bright light source approach to stray light testing reveals other forms of stray light (not limited to veiling glare). Dynamic Range  a comprehensive introduction: strongly affected by flare light. Contrast Resolution – A special transmissive chart for measuring the visibility of low contrast features in larger fields over a wide dynamic range Making Dynamic Range Measurements Robust Against Flare Light

 

Introduction to veiling glare  |  Lens reflections  |  “Black hole” light trap  |  I. Stepchart measurements
II. ISO 9358  |  III. ISO 18844  |  Gamma recommendations  |  IV. Contrast Resolution two-exposure method
Quantifying Veiling Glare  |  Two image calculation 

Introduction

Veiling glare is stray light that fogs images (reduces contrast in shadow areas). It is caused by reflections between surfaces of lens components and the inside barrel of the lens. It is a strong predictor of lens flare— image fogging (loss of shadow detail and color) as well as “ghost” images— that can degrade image quality in the presence of bright light sources in or near the field of view. It occurs in every optical system, including the human eye. It is a major factor in limiting camera dynamic range, especially when High Dynamic Range (HDR) sensors, which have sensor dynamic ranges of 120 dB or more, are used.

Veiling glare can only be measured reliably from linear RAW images, converted to 16 or 48-bit files with gamma = 1. Processed images (JPEGs from cameras, etc.) handle dark tones (critical to veiling glare measurements) in unpredictable ways that can affect results.

Lenses with low flare have been traditionally known for their excellent color performance. Color saturation is higher, especially in shadows. Low flare may be as responsible as sharpness for the exalted reputations of classic Leica and Zeiss lenses.

This statue near the entrance to Parc du Cinquantenaire in Brussels (the location of the autumn AutoSens conference) knows how to deal with flare light by shielding her eyes from the sun.

There are several ways of dealing with lens flare.

• In lens design, veiling glare is controlled by using high quality lens coatings (multi-layer coatings are best), baffling in the lens barrel, and careful design.
• In the field it is mitigated by lens hoods and anything that can shield the lens surface from the sun. Ansel Adams used a bellows lens shade, made famous in a Datsun (Nissan) commercial. (Here is some background information. I remember Adams’ Ford with “Zone V” license plates). He used his hat in a pinch.
• In the studio it is controlled by “barn doors” on light sources.

Veiling glare can be measured by the following Imatest modules.

• Stepchart uses a Q13 or Q14 (linear) grayscale chart adjacent to the black hole. The grayscale chart is used to estimate the camera response (for processed images). It is always more reliable to use raw images.
• Uniformity (and Uniformity Interactive) measure veiling glare using ISO 18844 method C in Imatest 5.0+. the image cannot contain a grayscale chart (linearization information must come from separate measurements.) Methods A and B, which require two chart images, will be added later.
• The Image Statistics module is valuable for examining spatially-dependent lens flare, as illustrated below.

Obligatory bad photo showing severe lens flare caused by the sun (far left), and moderate flare with the sun blocked (near left).
Canon EOS-20D, 24-70mm f/2.8L, 24mm, f/8, no hood.

Taken during Colorado’s notorious “Winter of 2007”
(five feet of snow in three weeks in December and January)

Imatest veiling glare measurements are designed to give a good prediction of a lens’s susceptibility to lens flare in challenging situations: the larger the number, the worse it’s likely to be. But it’s not a perfect measurement. It can’t predict ghosting (visible behind the Stop sign in the image on the left). And it is not reliable for processed images; only linear RAW images give reliable measurements. And RAW images are often not available for inexpensive cameras and camera phones.

The measurement is fairly simple to perform, but several pitfalls can cause inaccurate or misleading results unless measurements are made with care. Some degree of exposure control— either a manual setting or exposure compensation— is recommended for best accuracy. Exposure control may not be available with camera phones and other simple devices.

Veiling glare is measured by photographing one or more perfectly black regions inside a uniform white field that extends well beyond the image frame (1.41x the image width in ISO 18844). It is defined as the ratio of the light reaching the sensor from the black region (which emits no light) to the light reaching the sensor from the surrounding white field.

Veiling glare = V = L(black region) / L(white surface), where L is the illuminance at the sensor.

For linear image sensors (which includes most non-HDR sensors), this measurement is straightforward

• if you have access to the linear (RAW) output of the sensor,
• if the white area is unsaturated (ISO 18844 recommends a level of 225 out of 255 for files with bit depth = 8),
• if the signal in the black region is large enough to be well above the noise, and
• if the black region is perfectly black.

We deal with the last of these issues first: how to create a perfectly black object— a black hole (also called a light trap). Don’t worry: it’s more environmentally friendly than the type that sucks planets into its core.

Lens reflections are a major cause of flare light. An uncoated glass surface (index of refraction ≅ 1.5) reflects R = 4% = 0.04 of the light incident on it. (Remember, a sheet of glass or a lens component has two surfaces.)  For each glass surface between any surface and the light source, a fraction R of the primary reflection (R2 of the original incident light) is reflected back to the image sensor. This are called a secondary reflection. Since most lens surfaces are curved, this light will be unfocused, i.e., it will tend to fog a portion of the image.  According to Edmund Optics, the best anti-reflective coatings have R ≅ 0.4% = 0.004 over the visible spectrum (~400-700nm). R = 0.005 may be more realistic for a reasonable range of incident angles. The light reflected back to the sensor from each secondary reflection would be R2 = 0.000025 = 2.5*10-5 = -92 dB (20*log10(R2)). The number of secondary reflections Nsec increases rapidly with the number of components M (groups of elements cemented together, each of which has two air-to-glass surfaces) in a lens: 1 for 1 component; 6 for 2 components; 15 for 3 components; 28 for 4 components; 45 for 5 components, etc. For M components,      \(\displaystyle \text{Number of secondary reflections} = N_{sec} = \sum_{i=1}^{2M-1}i = 2M(2M-1)/2 = M(2M-1))\) M = 5 components are typical for high quality camera phones; M ≥12 components is commonplace for DSLR zoom lenses. Overall lens flare is less severe than the number of secondary reflections suggests because the stray light don’t cover the whole image. It decreases with distance from bright regions. It’s easy to see why practical camera Dynamic Range measurements are limited to around 70-90dB, even when sensor Dynamic Range is much higher. Image Statistics cross-section of a Contrast Resolution image for a high quality automotive camera, showing spatially-varying flare Because the ISO 18844 model does not measure the spatially-dependent flare caused by lens reflections, it’s value in characterizing practical system performance is limited.

 

III. ISO 18844 flare (veiling glare) measurements  with Imatest Uniformity or Uniformity Interactive

ISO 18844 flare measurements have been added to the Uniformity and Uniformity Interactive modules in Imatest 5.0+. Currently, only Method C, which uses a single image where the dark areas are “black holes” (light traps) has been implemented.

ISO 18844 chart arrangements: either a single rectangle with dimensions 1/10 the image height and width (H and W)
or a set of circles arranged on a diagonal.

Imatest provides a backlit version of this test target. Transmissive charts should be used in a dark environment, taking care to minimize light reflected back to the chart.

The current ISO 18844 standard has one rather serious problem. In section 4.3.1 it gives an equation for linearizing the image (finding the output luma value Y’sYCC) using the equation for the sRGB color space, which is used in most consumer cameras. The problem is that most cameras do not encode their images according to the sRGB (or any other color space) standards. In fact they’re rarely even close. Most cameras apply tonal response curves (with “shoulders” to soften highlight saturation), and these curves may also affect response in the darker areas (critical to veiling glare/flare measurements) that can be difficult to measure.

Sample ISO 18844 flare image (simulated). Click to download

• The first steps— illuminate the target through exposure and RAW settings are the same as for Stepchart.
• Photograph the central portion of the target, framing it for H and V in the above diagram, making some adjustment for aspect ratio if necessary.
• Run Imatest and select Uniformity or Uniformity Interactive. Open the image file. If the image is the same size as the previous run you’ll be asked if you want to repeat the same ROIs (Regions of Interest). In most cases you’ll select the entire image.
• After you’ve made the selection, if you click Yes (not Express mode) the Uniformity settings window, a portion of which is shown below, appears. You can select between no ISO 18844 calculation or the calculation with two approaches to linearization: enter the gamma (shown below to the left) or use the color space linearization (called for in the standard, but rarely correct). Gamma must be measured in a separate image.

Uniformity settings window (portion)

Gamma recommendation — To obtain reliable flare measurements, an accurate measurement of gamma in deep shadows (where flare is measured) is required, and such a measurement is difficult for several reasons. Although encoding gamma is specified by the color space (sRGB has a linear region in deep shadows), most cameras do not encode images according to the specification. They encode images to look good, and this means anything goes in deep shadows. Gamma in deep shadow regions cannot be extrapolated from gamma in lighter gray regions. For this reason we recommend measuring flare with gamma = 1 images converted directly from raw files. We do not recommend measuring flare with JPEG images obtained directly from cameras.

 

• Choose a “black hole” detection method: “auto” or “from manual threshold“
  • Auto (recommended): Applies adaptive thresholding to the image to determine what pixels constitute black hole areas. Less likely to be fooled by vignetting than a manually-set global threshold value.
  • Manual Threshold : When this option is selected, the field to the right is enabled and the user may enter a numeric value between 0 and 1. Linearized pixel values (normalized to the data type range, so that they area between 0 and 1) below this threshold value will be indicated as black hole areas.
  • Select the “Flare plot” to have the figure generate.

Results

Results are in the ISO 18844 flare plot and in the CSV and JSON output files.

ISO 18844 Uniformity results for a simulated chart (not all diagonal regions were included)

 

Black Hole Detection Problems

Imatest uses a few criteria beyond the thresholding option (auto or manual) described above to determine what is a valid area for producing ISO18844 flare results. If you are testing an image and the software appears not to be operating on a black hole area in it (i.e., it is greyed out in the output plot and shows no measurement numbers), there may be a few reasons why:

The black hole area is too small : In order to prevent spurious results, we require that areas that pass the thresholding test also have a large enough contiguous area to be useful. This involves eroding in from the sides of the detected area, so if your black hole is too small in the image (which may happen with low-resolution cameras at a large distance from the target), it may not appear in the results.

The black hole is not surrounded by enough light-area data : Method C of ISO 18844 requires that the pixel value in the black hole area be compared to the pixel values in the locally surrounding light areas. Moreover, it averages the light area values from regions from all four sides of the black hole (above it, to the left of it, etc.). If there is not valid light-area data within an appropriate margin on all four sides of the black hole area, it will not be measured. This can happen when the black hole is too close to the edge of the image/crop area or too close to another black hole. 

IV Contrast Resolution two-exposure method

In 2021.2 we added support for a new combined dynamic range and image flare (veiling glare) measurement made from two images of the Contrast Resolution chart: one acquired without a mask and one with a mask. In addition to veiling glare, this method measures two dynamic ranges: one with normal flare (no mask) and one with low veiling glare (masked), and provides additional information on the specific factors that limit dynamic range. The mask greatly reduces veiling glare by covering 86% of the lighter patches (1-12), leaving only the small lighter and darker inner regions. It also covers most of the remainder of the chart. Darker patches 13-20 are unmasked. An image of the mask, which fits tightly in the frame that holds the chart, is shown on the right, below. The circular openings (for the registration marks) are covered by neutral density gel filters with an optical density of 1.2 to minimize flare and (in most cases) allow automatic region detection. 

 

Contrast Resolution concept chart

Contrast Resolution flare mask

 

Method

Acquire two images of the Contrast Resolution chart, identically exposed and framed, one without the mask and one with it.

Exposure—  For linear sensors, the exposure should be set (if possible) so the lightest inner square in patch 1 is saturated, but the darker inner square is not. We will add a recommendation for HDR (nonlinear) sensors when we gain more experience. Manual exposure control is strongly recommended. Autoexposure should be avoided because most autoexposure algorithms tend to overexpose the brightest patches in masked images, often severely.

Framing—  We generally don’t recommend filling the frame with grayscale/dynamic range test charts. (Filling the frame is only appropriate for multi-purpose charts like SFRplus or eSFR ISO.) A good guideline for medium to high-resolution cameras (≥5MP) is that the vertical dimensions of the chart should cover about half the image.  Since veiling glare is dependent on the spacing between the lightest and the measurement patches, framing In any test sequence should be consistent to ensure measurement consistency.

Imatest automatically detects the mask and adjusts the calculations accordingly

The signal level, which is calculated from the left side of the lighter patches (1-12) in unmasked images, is replaced by the mean of the two smaller (lighter and darker) inner squares in masked images. The noise is calculated from one of the two squares (the lighter square, except for the top row, which can saturate). Most of the results are similar with and without the mask, except in the darkest regions. Since the mask greatly reduces veiling glare, the mean of log pixel levels in the dark bottom row (patches 18-20) is reduced. Veiling glare is the difference in the log pixel levels in the bottom row measured without and with the mask.

Results

We illustrate the two-exposure method with images from the Panasonic Lumix LX-100— a relatively compact high quality consumer camera with a 1 inch (linear; non-HDR) sensor and a high quality Leica-branded zoom lens that can save raw files (vital to the success of this measurement). Exposure control was manual, 1/60 sec., f/8. Three types of file were analyzed

• Raw files were converted to 48-bit color TIFF files (bit depth = 16) using LibRaw (the replacement for dcraw), which has no sharpening or noise reduction. Two gamma settings were analyzed.
  1. files converted to Adobe RGB color space with encoding gamma = 1/2.2, and
  2. files converted with gamma = 1 (unchanged from the raw image).
• JPEGs (24-bit sRGB color), generated inside the camera. Noise reduction and sharpening (bilateral filtering) are applied, and the tonal response curve is quite different from a straight gamma curve (pixel level proportional to brightness(encoding gamma)).

Raw to 48-bit color Adobe RGB (bit depth = 16; encoding gamma = 1/2.2)

This conversion gave the best results. Using gamma = 1 (described below) causes measurements to be limited by bit depth = 16, which has a maximum pixel level of 2¹⁶-1 = 65535, where log₁₀(65535) = 4.816 = 96.3 dB. As we point out in Gamma, Tonal Response Curve, and related concepts — Advantages of gamma-encoding, gamma-encoding (typically with γ = 1/2.2 = 0.454) significantly increases the effective dynamic range, greatly improving these measurements.

This conversion encodes the input image with a nearly pure gamma = 1/2.2 curve. If there were no veiling glare or noise, the response curve (the thick gray line in the plot below) would be a straight line with slope = 1/2.2 all the way from the lightest to the darkest patch. A flattening of this curve in dark regions indicates the presence of veiling glare (and possibly a noise floor).

The key results are on the right side of the upper graphs of each image (the font is somewhat small in the images below): Y(patch 1, mean(18-20)) is the log₁₀(luminance) level for patch 1 and the mean of the dark patches in the bottom row. This value is strongly influenced by flare light, as indicated by the flattening of the log₁₀ signal level curve in the upper plots.

Adobe RGB (48-bit) results without mask.
Patch 1 = 0; mean Y(patch 18-20) (bottom row) = -2.04 (log₁₀ units) Adobe RGB (48-bit) results with mask.
Patch 1 = -0.027; mean Y(patch 18-20) (bottom row) = -3.53 (log₁₀ units)

Quantifying veiling glare

Since the density of the gray areas increases by 5 dB = 0.25 OD per patch, the mean density of the bottom row (equivalent to patch 19) is 18×5 = 90 dB = 4.5 OD, equivalent to patch 19 illumination = 0.0000316 (linear units) relative to patch 1 (the brightest patch on the upper-left).. For a γ−enc = 2.2 color space, this would result in P_log(19) = log₁₀(pixel level(19) ⁄ p_max) = 4.5/γ−enc = −2.04, which is higher than we often observe. If P_log(19) is lower for either of the images (it’s -2.04 and -3.56 for the images above), we assume that the camera is not sufficiently linear to calculate veiling glare from a single image, and we have to use the 2 image method, below. 

The measured log pixel level for normalized illumination I_n on patch n is affected by the encoding gamma γ-enc.

pixel level(n) ⁄ p_max = I_n ^γ-enc   for patch n,   where p_max = the maximum pixel level for the bit precision (255 for 8-bit files; 65535 for 16-bit files, etc.)

For Adobe RGB color space (used here), γ-enc = 1/2.2 = 0.4545. sRGB is close but not exact (it has a small linear region.) Note that in-camera JPEG files apply all sorts of tonal response curves, making γ-enc difficult to determine. For γ-enc to be a reliable, known value, you must start with raw images and use a raw conversion program (such as LibRaw, used in Imatest) that uses a simple gamma curve without any additional processing like “shoulders” or tone-mapping. Log pixel level, displayed on the plots, is

P_log(n) = log₁₀(pixel level(n) ⁄ p_max) = γ-enc × log₁₀(I_n) ;          I_n = (pixel level(n) ⁄ p_max )^1/γ-enc = 10^(P_log(n) ⁄ γ-enc)

For patch 1 (the brightest patch on the upper-left) we can assume that I is entirely from the chart (I_1-total = I_1-chart ), but for the mean of patches 18-20 (i.e., patch 19), where the chart has an average density of 4.5, corresponding to a level off 0.0000316 (linear units) relative to patch 1 (I_19-chart = 0.0000316 I_1-chart), we assume that I_19−total is the sum of illumination from the chart and from veiling glare,

I_19-total = I_19-VG + I_19-chart = I_19-VG + 0.0000316 I_1-chart ;        I_19-VG = I_19-total – 0.0000316 I_1-total

We don’t want an absolute number. We want the veiling glare measured on the mean of patches 18-20 relative (normalized) to the illumination on patch 1. For brevity we call it VG₁₉.

VG₁₉ = I_19-VG / I_1-total = I_19-total / I_1-total  – 0.0000316  

VG₁₉ is calculated in several steps from the above equations, starting with P_log(1) and  P_log(19), using I_1-total and I_19-total as intermediate calculations. It should be a reasonably good number for linear sensors where γ-enc is known and consistent, but we have not found it to be reliable in practice. For the masked image above, I_19-total is much lower than expected for patch 19, assuming good linearity and γ-enc = 1/2.2. For this reason measurements with 2 images are more reliable.

Two image calculation — Because camera linearity may not be what we expect (as we found for the above example), a better estimate of veiling glare is the difference in VG₁₉ from two images, where we ignore the expected 0.0000316 value ffrom patch 19. VG₁₉ = I_19-total / I_1-total . 

VG₁₉(2 images) = VG₁₉(no mask) –VG₁₉(masked) 

 = I_19-VG(no mask) / I_1-total(no mask) − I_19-VG(masked) / I_1-total(masked) 

 If With manual exposure control (required for good results),  I_1-total(no mask) ≅ I_1-total(masked), but it’s easiest to measure them in the individual images.

For the image with no mask, 

The difference in the results — -2.04-(-3.53) = 1.49 log₁₀ pixel units — is an indicator of veiling glare, though it’s the primary measurement. Since the image was encoded with gamma = 1/2.2 (0.4545) (close to the measured value of 0.458), the actual flare difference is 1.49×2.2 = 3.28 density units. Note that the relative mean level of the patches in the bottom row is log10^-3.53 = 1/3388.

A little more math The nominal density for each of the twenty large patches in the Contrast Resolution chart increases by 5 dB (0.25 density units) per patch. The darkest patch is 95 dB (4.75 density units) below the lightest, equivalent to 1/56234 = 0.00001778. If a system were exactly encoded with gamma = 1/2.2 (perfectly linear in log space), the darkest patch would have log10(pixel level) = 4.75/2.2 = 2.16. Since this is not what we measure, we have to conclude that the system (hardware or LibRaw software) does not follow the gamma = 1/2.2 curve exactly in the dark regions. The mean luminance (Y) values for the bottom row without the mask in linear units is 10-2.04 =  0.00912 = 1/109.6. The mean luminance (Y) values for the bottom row with the mask in linear units is 10-3.53 =  0.000295 = 1/3388. This level is due to signal from light passing through the chart, noise, and veiling glare from the bright part of the masked image (much less than for the unmasked image). A good measure of veiling glare is (the level of the darkest row in the unmasked image minus the level in the masked image) divided by the level of patch1, which should be exposed close to saturation. If we assume that these numbers are additive, we can subtract the masked level to find the veiling glare level. Veiling glare level = 0.00912 – 0.000295 = 0.008825 = 0.8825% (linear units) = -2.05 (log10 units).

Raw to 48-bit color (bit depth = 16; linear encoding: gamma = 1)

The gamma = 1 results appear similar to the gamma = 1/2.2 results (except for the larger slope: measured gamma = 0.847), but the numbers were different. Mean Y (patches 18-20) = -3.86 without the mask and -5.32 masked. The difference between the two results, 1.46 density units, is similar to the difference for gamma = 1/2.2, but this is less than expected. Based on the gamma = 1/2.2 results we would have expected a difference of around 1.46×2.2 = 3.2.

What appears to be happening is that log₁₀(pixel level) drops below -5, which is 10^-5 = 1/100,000 in linear units. This is beyond the expected range for 16-bit depth images, where the maximum pixel value is 65535, i.e., the limit of the linear 16-bit depth file has been reached. The histogram on the right shows pixel levels in a dark portion of the masked image. Many pixels, particularly in the green channel, are zero. Note that this is for a 16-bit file, which has a maximum level (x-axis value) of 65535.

JPEG (24-bit color; bit depth = 8; encoding gamma approximately 1/2.2)

JPEG files are processed inside the camera. Processing includes noise reduction (which accounts for the larger noise-based dynamic range than the 48-bit Adobe RGB conversions), sharpening, and the application of a tonal response curve in addition to gamma-encoding. The tonal response curve is not straight: it has a strong “shoulder” at bright levels, which makes the image appear more “film-like” and helps to minimize saturated (burnt-out) highlights.

For both the unmasked and masked JPEG images, the response curve flattens out around input density = 3 to a log₁₀(level) of around -2.3 (10^-2.3 = 1/200. Mean Y (patches 18-20) is -2.19 without the mask and -2.24 with the mask. Sad to say, the 2-image veiling glare measurement fails for the 8-bit JPEG image. :-(  The 16-bit Adobe image (encoding gamma = 2.2) works best.

JPEG (24-bit) results without mask.
Patch 1 = 0; mean Y(patch 18-20) (bottom row) = -2.19 (log₁₀ units) without the mask
and -2.24 with the mask. The measurement failed because of limits of the 8-bit image.

 

A deeper dive into veiling glare: why the bottom row reverses direction

A strange anomaly is visible in the bottom row (patches 18-20) of the above images: log density starts increasing, when it should be decreasing. To observe the deep shadows, we opened a heavily overexposed capture of the masked chart in Image Statistics (1 sec. vs. 1/60 sec. nominal exposure: 60× the exposure), then displayed it using Lighten (HSL) to makes deep shadow detail visible. The 7 lightest patches are fully saturated.

Short-distance flare, as described in the Dynamic Range post and measured by ISO 18844, is plainly visible.

Something surprising but not entirely unexpected is happening in the bottom row: smudges that get lighter as we move from patch 18 to 20 (from dark to darker). These smudges appear to be radially-symmetric highly-blurred replicas of the light patches in the first row, indicated by the red arrows pointed from the source of the smudge to the smudge itself. This phenomenon is uncommon, though we’ve seen it in one other camera.

It is interesting to compare it with a 30× overexposed unmasked image, also HSL-lightened. Lots of veiling glare is visible.