What is the advantage of obtaining 3 bands (RGB) imagery with 16 bits pixel depth?
In terms of visual context, is an image of 3 bands (RGB) with 16 bits pixel depth necessarily better than the same image with 8 bits?
Is the pixel depth value set at the level of camera? how about the number of bands?
Answer depends on the sensor and processing applied. Most modern satellite and aerial sensors capture 12-14 bits per pixel enabling a wider dynamic range to be captured and reducing the quantization of the imagery. The larger bit depth enables more details to be captured especially in dark and light areas of the image. This is only applicable if the sensor also has low noise. The value of higher bit depth is also more evident in images that have more significant light fall off or other trends in the image. If the image has been correctly processed vast majority of the information content will be within the 8bit range. Use of 16bit imagery is typically used for imagery directly from the sensors. The number of bands depends on the sensor. Note some sensors such as those in most drones actually only have one band in the raw data, but include filters (Bayer) on the CCD/CMOS that then enable the interpolation of a 3band image. Most images from such cameras are recorded as 3band JPEGs, although some cameras can also record the RAW sensor data, which can be post processed to gain a bit more detail. Some more advanced sensors capture 4 bands typically to include near Infrared bands for agricultural applications. There are also cameras and sensors that capture more bands for specific applications. For the majority of application dealing with the visual interpretation of processed imagery 8bit 3band imagery is sufficient.
Thank you very much Peter for the useful input.
In our case, the purpose of capturing the imagery is the visual context.
- Do you recommend us to go for 16 bits or 8 bits (with 3 bands in both cases) knowing that our interest is exclusively on how the image is visually clear?
- By how much the imagery with 16 bits will be clearer (in terms of visual context) than the one of the 8bits (both with 3 bands)?
- By how much the storage size of the imagery with 16bits is bigger than the one of 8bits (roughly)?
If the imagery has been correctly processed then 8bit imagery will provide the vast majority of the details. If you are capturing imagery directly from a sensor I would use 16bit. There will be no visual difference until you start further enhancing the imagery. ArcGIS provides features such as DRA (Dynamic Range Adjustment) that enable the full details such as shadows and highlights to be seen and for this 16bit can provide more information.
Assuming no compression 16bit takes 2x data volume of 8bit. With compression these factors change. JPEG can be used to compress 8bit imagery very well and can achieve about 8x compression with nearly no visual loss. Esri support also 12bit JPEG compression can can be efficiently used to compress imagery where the data range has been set to 0 -4096. This is often a good way to get additional dynamic range while enable compression of about 6x. For 16bit imagery one can use JP2 which can get about 30% additional compression, but I would not recommend due to the significant conversion required. Esri also support LERC which can typically get about 2.5x compression of 16bit imagery losslessly and by allowing some loss in precision the compression can be greater, but for optical imagery I would use JPEG if size is a significant concern. For more details on this see Imagery Formats and Performance
Thank you for the prompt answer.
To give better context for our issue:
We need to decide the optimal choice for capturing aerial photos out of which orthophoto is derived. There has been a big discussion pertaining to the available options considering that fact that purpose of the required orthophoto is limited to the visual data extraction (no remote sensing is required). The clearer (visually) the orthophoto the better. In this case:
Is it better to go for 16 bits pixel depth with 3 bands or 8 bits with same number of bands (the GSD in both cases is 5cm)? does the 16bits increase the visual context twice in comparison to the 8bits?
If you are referring to the imagery being captured by the sensor then I would recommend 16bit 3band. Ensure you use the OptimizeRaster tools to structure them appropriately. I would recommend you use MRF with LERC compression for best lossless compression, but you may prefer to use Tiled TIF Deflate. You can use these in ArcGIS for example using Ortho Mapping to generate the ortho images. If you are using the dynamic mosaicking capabilities of ArcGIS to gain the maximum detail, then leave the images as 16bit. If you are looking for a single seamless image then use the color correction (dodging) and seamline generation then export (as TileCache, CRF or TIF files) but set the output to 8bit and use a JPEG quality of 95%. This will significantly reduce the data volume and provide the visual product that you are most likely looking for.