FIXME **This page is not fully translated, yet. Please help completing the translation.**\\ 
// (remove this paragraph once the translation is finished) //

====Noise====

**Definition and Sources of Noise**\\

In the field of imaging, **noise** typically refers to random interference signals carrying non-target information introduced during image acquisition, transmission, or processing. It manifests as irregular variations in luminance or color (such as graininess, color mottling, or irregular spots), directly affecting image clarity and information reliability.

**Sources of Noise:**\\
One type originates from the physical characteristics of the light source (photon shot noise), while the other arises from the limitations of sensor technology (dark current and read noise).\\
{{ http://ydadmin.rdbuy.com.cn/ueditor/php/upload/image/20230809/1691567494160176.png }}\\

**Signal-to-Noise Ratio**\\

Signal-to-Noise Ratio (SNR) is a common metric for describing noise. It refers to the ratio of signal to noise in a system, often expressed in decibels (dB). For example, imagine a large room with many people; two individuals standing at opposite ends want to communicate. The content of their communication is the signal, while the voices of others around them constitute the noise. The SNR value indicates the impact of the surrounding noise on the communication; they must find a way to overpower the ambient noise to communicate effectively. Therefore, in image processing, a higher SNR means a cleaner image with minimal noise interference (such as grain or snow-like artifacts), resulting in a visually pleasing image. Conversely, a low SNR may result in an image filled with snow-like artifacts, severely degrading image quality. To measure the SNR of an imaging device, one can refer to the ISO 15739 standard. But is this the end of the topic? No, it is not. As a method for describing noise, SNR does not adequately reflect how much noise an observer can actually perceive in an image.\\

We have been measuring digital cameras since 1997 and used SNR to measure noise for a long time. However, over five years ago, SNR measurements no longer accurately reflected the noise perceived by human observers. Cameras with similar SNR values can exhibit different visual noise appearances. The same perceived noise can result in different SNR values.\\

Next, we use three example images: 1x, 2x, and 4x to illustrate this issue.\\

{{ http://ydadmin.rdbuy.com.cn/ueditor/php/upload/image/20230809/1691567537429531.png }}

These images are supposed to be uniform, but they exhibit some noise. If you step back from the screen, you can see that the noise in all images appears roughly the same and looks uniform. However, if you get very close to the screen, you can see that the 4x image has more noise than the 1x image.\\

If we measure the SNR of these three images, we can see that the average values of all images are the same, all pixel values have the same standard deviation, and thus the same SNR values.\\

{{ http://ydadmin.rdbuy.com.cn/ueditor/php/upload/image/20230809/1691567564953302.png }}\\

This example illustrates that SNR only reflects the total amount of noise, but fails to reflect the noise perceived by human observers.

**Visual Noise**\\

Therefore, to describe how much noise observers can see in an image and whether it is disturbing, the concept of Visual Noise was introduced. The Visual Noise value is easy to understand: the higher the value, the more noise the observer sees. The main difference between SNR and VN is that VN weighs noise based on its visibility; noise that is completely invisible is not considered in the VN measurement.

{{ http://ydadmin.rdbuy.com.cn/ueditor/php/upload/image/20230809/1691567611479778.png }}

**How do we know which noise is visible and which is not?**

The response of the human visual system to spatial frequencies can be modeled. The Contrast Sensitivity Function (CSF), along with assumed viewing conditions, allows us to calculate the importance of different parts of the noise spectrum. Therefore, from the above example: Image 1x has significant noise at high spatial frequencies, where the CSF has a low response. Image 4x has most of its noise at low spatial frequencies, which is well observed through the CSF. Consequently, the Visual Noise value of Image 4x is higher than that of Image 1.

{{ http://ydadmin.rdbuy.com.cn/ueditor/php/upload/image/20230809/1691567636718657.png }}

**How to Measure Visual Noise?**\\
The process and algorithms follow ISO 15739. The simplest method is to use an OECF chart (here, our self-developed CP197-TI transmissive grayscale test chart is used) and RIQA software. A VN value is obtained for each channel, and these are finally connected to form a line chart, as shown in the gray area of Figure 6.

|  [[https://www.yanding.com/product/detail?id=1571|{{ http://ydadmin.rdbuy.com.cn/ueditor/php/upload/image/20230809/1691567655225960.png }}]]  |
^  [[https://www.yanding.com/product/detail?id=1571|Figure 5: CP197-TI Transmissive Grayscale Test Chart]]  ^

|{{ http://ydadmin.rdbuy.com.cn/ueditor/php/upload/image/20230809/1691567680157428.png }}|
^  Figure 6  ^

**See Also**\\
[[:camera测试用例]]、[[:测试用例]]、[[:riqa简介]]