Image File Formats

Signal intensities are encoded as numbers in the image file, one number for each pixel. For color images, things are a bit more complicated: for each color channel, one number has to be stored, for example one intensity each for red, green, and blue. Make sure you use grayscale instead of color images: the extra color information provides no benefit for the subsequent analysis of 2D gels.

When your image file is, for example, a "16-bit TIFF" file this means that image intensities are encoded with 16 bit numbers, giving 65,536 (2 to the power of 16) possible different values for each pixel. In contrast, an 8-bit image file only stores 256 different values per pixel. The number of bits per pixel is also called the color depth. The following table shows some examples.

TIFF images can have different color depths, this is one of the reasons why TIFF is a widely used image file format.

Generally, having more possible intensity values per pixel (higher color depth) is better for the analysis. The tradeoff is between higher accuracy and need for more space to store the information: a 16 bit TIFF file is twice as large as the equivalent 8-bit file, but gives you 256 times more nuances in the image. Another consideration is that many everyday image processing programs are not able to deal with color depths greater than 8 bit.

Decreasing color depth in the scanned image file normally results in loss of accuracy. It is not recommended to increase color depth after the scan is completed: if your scanner produced only an 8-bit image you have at most 256 different intensity values in the image. Converting the file to a 16-bit image will only give you at most 256 out of 65,536 possible gray values.

You can find more information about image file formats in the Wikipedia Category: Graphics File Formats. At the ProteomeInformatics.net Proteomics Image Analysis Forum there is a discussion thread "Compatibility of Image formats across different Image Analysis Softwares" that focuses on experiences from the proteomics community with vendor specific formats.

Data reduction and image calibration

Some imaging devices can measure more intensity values than what fits into the available image formats. One way to deal with this is distributing the intensity values linearly over the whole intensity range that is offered by the image file. An example: Say, the Aanalog to Digital signal converter can deliver 1024 intensity levels, but the image file is limited to 256 levels (8-bit color depth). Linear transform condenses 4 A/D converter levels to 1 intensity level in the image. Of course, this process effectively wastes accuracy.

Especially if light scanners are used for image generation this technique can be improved by using only the real dynamic range delivered by the A/D converter. For our example this could mean: The A/D converter can deliver 1024 different intensity levels but in our experiment the gel only has intensity values in the range from 128 to 920. The 792 intensity levels have to be encoded in the image file - so only 3 A/D converter levels have to be condensed to 1 image intensity level. Compared to the former approach that wastes the lower and higher dynamic range this improvement results in a 25% finer resolution.

There are instruments that can distinguish 100,000 and more intensity levels, far more than can be encoded in a TIFF or similar file format. In order to save as much information as possible in these files, intensity levels are encoded using a nonlinear calibration curve. During scanning, measured intensities are converted to pixel gray values according to this curve. During quantitation, the image analysis software has to decode the pixel values to arrive at the originally measured intensities. The curve is designed such that lower intensities will be encoded with higher accuracy than higher intensities.

Example image calibration curve.

During image analysis, it is important that the image analysis software recognizes the calibration curve so it can do quantitation using the intensities originally measured by the scanner. General image processing packages such as Photoshop ignore grayscale calibration. It is even possible that calibration information is lost during processing with these packages. Additionally, TIFF files do not include calibration information so you will have to use vendor-specific formats.

File compression

Compression of image data means the usage of algorithms to reduce the size of a gel image file while retaining all or most of the image information. If you use calibrated image file formats then image compression is not an option because compressed file formats do not store the calibration information.

Image compression algorithms can be classified as lossy or lossless methods. Lossy compression means that information can be lost in the image details, giving a higher compression ratio than lossless methods.

Uncalibrated files may be saved in file formats that use loss-less data compression e.g. subformats of *.tiff / *.tif or *.png. Often compression to 50 % of the original size is possible. The commonly used JPEG format uses a lossy compression method. Large compression ratios may heavily change or destroy your data. For low compression ratios, there is still an unknown influence of compression on spot quantities. That is why we recommend to avoid using the JPEG file format for image analysis purposes.

Image compresson only saves disc space but does not affect the amount of working memory (RAM) needed for image analysis because compressed files will also be completely extracted before starting analysis.

The following table summarizes the properties for commonly used image file formats.