Skip to content
行业
产品
测试实验室
培训与活动
资源
公司
联系我们
联系我们

位流图像的转换

技术笔记

How RAW is "RAW"?

Understanding RAW Image Data: How unprocessed is a RAW file?

Introduction

Cameras that address experienced users, such as D-SLR and system cameras, normally offer a RAW-Mode. In this mode, the camera puts files onto the memory device, that are called “Raw”-Files, as they contain the pure data which was obtained from the sensor and has not been processed.

At least that is what these files are considered to be. The important question is:

At a Glance

  • RAW files are not always pure sensor output. Although RAW formats are intended to contain minimally processed image data, manufacturers often apply corrections and optimizations before saving the file.

  • Different cameras process RAW data differently. Measurements of camera response curves (OECF) show that identical scene information can be modified in different ways depending on the manufacturer and camera model.

  • Noise reduction can affect RAW files. Features such as dark-current compensation and long-exposure noise reduction may alter RAW data before it is written to the memory card.

  • Hidden processing may occur even when image enhancements are disabled. Experimental testing shows that some cameras still apply undocumented modifications to RAW files, even when noise reduction settings are turned off.

  • Comparing camera quality using RAW data is challenging. Since every manufacturer defines its own RAW processing pipeline, the "rawness" of RAW files varies, making direct comparisons between camera systems difficult.

Key takeaway: RAW files provide significantly more original sensor data than JPEG images, but they are not necessarily untouched sensor output. The question remains: How raw is RAW?

“How raw is RAW?“

In our experience, we can say: The files are as raw as the manufacturer wants them to be.

Some examples:

In the iQ-Analyzer, we can extract the data from the RAW files and use them for our measurements. Let’s take a look at the OECF (Opto-Electronic Conversion Function) of two different cameras (Canon 7D and Nikon D7000).

We see that the OECFs are different. One camera shows values close to zero in the dark areas, the other one does not. So does one of these cameras have no dark current? I don’t think so, most likely the dark current has been compensated in one camera and not in the other. So RAW is not the pure sensor output, it has already been modified in its tonal range.

The OECF (Opto Electronic COnversion Function) does not start from 0, even though luminance approaching zero. This is a typical behavior of a sensor, which generates some signal also without light; this is called dark current.
The OECF starts from zero in this example, we can assume that the sensor will have some dark current, but this is corrected before the data is written. So the output is not as the sensor has provided, but corrected by the dark current. A common technique, but causes differences between different RAW formats if applied or not.

Another example:

In the iQ-Analyzer, we can extract the data from the RAW files and use them for our measurements. Let’s take a look at the OECF (Opto-Electronic Conversion Function) of two different cameras (Canon 7D and Nikon D7000).

We see that the OECFs are different. One camera shows values close to zero in the dark areas, the other one does not. So does one of these cameras have no dark current? I don’t think so, most likely the dark current has been compensated in one camera and not in the other. So RAW is not the pure sensor output, it has already been modified in its tonal range.This test has been performed with a Nikon D3x. The images have been taken in complete darkness with a really long exposure time (several minutes) to show the effect of dark current.

The user can decide if a dark current compensation for long exposure times is wished or not. So if we turn the compensation off, we get colorful images, which show us the noise due to the dark current.

If we turn the compensation on, the camera automatically starts a second exposure right after the first one, without opening the shutter and lasting just as long as the first one. The obtained image is used to correct the first one. The result is a significant lower noise; the corrected image is nearly black.

Interestingly, we can have a third mode, which is not documented. To get this, we turn the camera off while it exposes the second image. So take an image and turn it off directly after the shutter has been closed. The camera will save the first image into the memory device without the correction with the second one. In this case, it shows even more noise compared to the images we have taken without the compensation. So this is an indicator, that also the images we have taken with compensation turned off, have been modified with some kind of processing.
To make sure that this is not only an effect in the JPEG files, we had a look at the RAW files and could make the same observation: The noise is different in the three images, so there has been some processing on the RAW files.

Long exposure time, noise reduction turned off.
Long exposure time, noise reduction turned on.
Long exposure time, noise reduction turned on, camera interrupted while taking the correction image.
Long exposure time, noise reduction turned off.
Long exposure time, noise reduction turned on.
Long exposure time, noise reduction turned on, camera interrupted while taking the correction image.

So we see: The files are as raw as the manufacturer wants them to be.

If a manufacturer leaves the image rawer than its competitors, this can mean worse noise results in the RAW file compared to another manufacturer who already somehow processes the files. A fair comparison and judgment of the quality of the camera is extremely hard to do on RAW, as you never know: How raw is RAW?

About this article:

Written by the Image Engineering R & D team. Technical review: Uwe, Artman. Last updated: 2026 . Image Engineering is an independent test lab for camera image quality and the developer and manufacturer of the complete test infrastructure, including test charts, analysis software, and measurement equipment.

准备好实现您的愿景了吗?

了解我们的精密测试解决方案如何提升您的成像项目。

需要帮助吗?

图像质量测试?

我们的专家团队随时准备为您提供帮助,助您在成像应用中获得最佳效果。

Phone (DE): +49 2273 99 99 1-0
Phone (USA): +1 786 686-8664
Phone (China): +86 158 8961 9096
Office Hours: Mon–Fri, 9–5 PM (CET)
联系我们
Image Engineering
带光圈的镜头 - 标准图像
A graph illustrating the relationship between luminance in cd/m² and digital values for different colors, with an additional line representing SNR.
A OECF graph showing digital values and SNR against luminance in cd/m². There are four lines representing R (red), G (green), B (blue), and SNR (pink dashed line).
Image noise in JPG 1
Image noise in JPG 2
Image noise in JPG 3
Image noise in a RAW file 1
Image noise in a RAW file 2
Image noise in a RAW file 3
Image Engineering Logo