Image Quality Factors

Texture Loss

Image Quality Factors
  1. Introduction
  2. How to measure texture loss
    1. Low contrast Siemens star
    2. What is a dead leaves target?
  3. How to analyze texture loss using a dead leaves target
    1. Dead leaves direct method
    2. Dead leaves cross method
  4. Conclusion
  5. References


The texture in digital imaging is often referred to as low contrast fine detail. For example, the individual blades of grass of a meadow or the leaves from a tree are the fine details or texture. The camera should be capable of reproducing these details. The technical terms used in this context are high spatial frequencies that often combine with low local contrast and color differences. For the camera, it is difficult to differentiate texture from the unwanted noise introduced by the camera at higher ISO sensitivity settings in situations with low light levels. So, a reduction of the noise level in an image often leads to texture loss.

Simply stated, texture loss is the loss of fine details in an image due to a reduction of noise (Figure 1). With today’s advanced image processing, it has become easier to reduce unwanted noise and other artifacts. However, this process can also diminish important texture leading to poor image quality.

Of course, it is recommended to find the right balance between noise reduction and texture loss. To do this, noise, as well as texture loss, must be measured and analyzed.

Figure 1: Texture loss of fine details due to noise reduction

How to measure texture loss

A texture loss measurement is technically a resolution measurement that uses a different test target as opposed to the ones described in ISO 122331 (the main resolution standard). During testing, the device under test will reproduce the test target and the analysis software (e.g., iQ-Analyzer) will provide an SFR (spatial frequency response). This SFR describes how well the system can reproduce the texture in the image.

We define texture as low contrast and fine details; thus, our test target needs to also show low contrast and fine details. Two different test targets are available for testing a camera for its texture loss.

Low contrast Siemens star

The ISO 12233 standard describes the s-SFR method based on high contrast, sinusoidal Siemens star (Figure 2). This method is known to be very robust against sharpening and it a great tool to obtain a reliable measurement of the limiting resolution2.

Meanwhile, ISO 195673 specifically addresses texture loss measurements. This standard is based on the s-SFR method but uses a low contrast version of the Siemens star (Figure 2).

siemens star chart
siemens star low contrast chart
Figure 2: A comparison between a normal Siemens star (left) and a Siemens star with low contrast (right)

In short, the analysis software will detect the center of the Siemens star and then read out digital values along a radius from the image. These values can then be used to calculate the modulation, which will result in an SFR if done for all radii and angles. So, the calculation of the SFR is fully based on an analysis of the digital values if we stay within the spatial domain.

What is a dead leaves target?

The dead leaves chart (Figure 3), as opposed to the Siemens star chart, creates a more natural testing scene or test structure for texture loss.

The chart itself was introduced as a model for natural images. Essentially, it is a large stack of circles with a known probability for their size and color. In its first appearance, the dead leaves target was still a gray target, while in the latest research4 it is used in a color version (Figure 3), as that better represents a more natural scene and ultimately leads to a better correlation between measurement and user experience.

dead leaves chart
dead leaves color chart
Figure 3: A comparison of the gray vs. color dead leaves target

How to analyze texture loss using a dead leaves target

Over the years, different methods for the analysis of an image showing the dead leaves pattern have been developed. The early ones were very much influenced by image noise and could be seen more like a “proof of concept” rather than a stable measurement.

The dead leaves target also looks more natural than other test patterns. Another benefit is its power spectrum that can be calculated from some simple properties during the creation process.

If we know the power spectrum of the target Y(f) and we measure the power spectrum in the image X(f), we can simply calculate the absolute of the transfer function H(f) as seen below (Figure 4).

dead leaves analyzation.png dead leaves transfer function
Figure 4: Calculating the absolute of the transfer function

Dead leaves direct method

With the work from John McElvain, a more robust and reliable analysis method is available. As in the earlier approaches, it calculates the power spectrum of the target and measures the power spectrum in the image. It also takes into account that the camera will not only remove spatial frequencies, but it will also add noise and thus high spatial frequencies as well. The dead leaves direct method adds the measurement of the noise power spectrum and uses it for a correction (Figure 5). For the noise measurement, an additional uniform gray area is required.

This approach is also used in the IEEE-P1858 CPIQ (reference link?) test procedure.

dead leaves direct calculation method
Figure 5: Calculating texture loss using the dead leaves direct method

The dead leaves direct approach takes into account the noise that the camera will add to the image and in turn affects the measurement of the power spectrum. While this approach is already much better than the earlier approaches, it still encounters a few problems.

When we measure texture loss, we have to assume that the camera more or less performs a noise reduction. A modern noise reduction algorithm will function differently depending on the image content. In essence, we know that the noise reduction depends on the image content, and that the noise added to the dead leaves structure by the camera is the same noise that is added to a uniform gray patch. However, this assumption is not valid in many cases and we actually measure less noise on the gray patch than we find on the dead leaves structure.

Dead leaves cross method

A completely new approach was introduced in 2014. This approach is called the dead leaves cross6 due to it using a cross-correlation between the image and the target data.

This method compares the image content with a reference image that needs to be calculated specifically for the image captured based on information about the target, e.g., the distortion, shading, and tone curve. If this approach is applied, then the complete transfer function can be calculated (FIgure 6) including the phase information, and then we can calculate the SFR without the influence of image noise or other artifacts.

dead leaves cross calculation method
Figure 6: Measuring texture loss using the dead leaves cross method


After we have performed the analysis, we will have an SFR. This SFR describes how well the device under test can reproduce the target. In other words, the lower the SFR, the more texture loss we have. As the lens and sensor will contribute to the SFR, we advise comparing the SFR obtained from the dead leaves to another robust method to measure the SFR. We recommend the s-SFR method, based on the high contrast Siemens star.

Test results and a comparison between the different methods is available on our website7. The test target used for these experiments is the TE42, analyzed by the iQ-Analyzer software.