Word of the Month – Bit Depth, Part II

For the July edition of Word of the Month (WOM), we continue the discussion started last month where we laid the ground work for understanding bit depth as well as explained the differences between 8-bit and 11/16-bit depth imagery. In this month’s WOM, we continue the discussion on bit depth focusing on how its potential impact your imagery ordering habits.

Before we jump into the practical applications of 8-bit versus 11/16-bit imagery, we should spend some time on the conversion of 11/16-bit imagery to 8-bit depth. Last month, we explained that all current high-resolution sensors collect raw satellite imagery with 11-bit depth; DigitalGlobe then adds extra zeros to create a 16-bit depth product while GeoEye delivers imagery with 11-bit depth. As such, we will drop the convention of typing ‘11/16-bit depth’ moving forward and simply refer to all this data as ‘11-bit depth’. As was explained in last month’s WOM, imagery with 11-bit depth has 2,048 possible values per band of imagery versus 8-bit imagery which only has 256 possible values per band. While the mathematical process of converting high-resolution imagery from 11-bit to 8-bit is outside the scope of this article, we can explain the basic concept behind this re-scaling. If you divide the number of possible values in 11-bit depth imagery by the number of values possible in 8-bit depth, you will end up with the number 8 which means that 11-bit values compress into 8-bit values in the following fashion:

From the illustration above, one can see that rescaling 11-bit imagery down to 8-bit depth is an irreversible process unless one saves a copy of the raw 11-bit data. Further, you can see that by rescaling from 11-bit to 8-bit depth, much of the subtle variation between pixel values in 11-bit imagery is lost. This loss of variation has ramifications on the uses of 8-bit versus 11-bit depth data.

For many of our clients, the most important information contained in high-resolution satellite imagery is the picture itself. These clients care little about the spectral information that lies behind this ‘pretty’ picture. For these clients, there is no advantage in purchasing 11-bit depth imagery and it does in fact add extra steps when using the data. When imagery is delivered with 11-bit depth, it is up to the user to color balance the data so that trees are green, sidewalks are white and roads are black. While this sounds like a relatively easy process, it can be challenging for novice imagery users and even experienced users. For those simply wishing to use their satellite imagery as a backdrop in their projects, eMap strongly suggests purchasing all your imagery products in 8-bit depth with the color balancing already applied. To this end, eMap offers a proprietary processing technique called ImageBoost which sharpens and applies a highly customized color balance to 8-bit high-resolution imagery.

Some of our more experienced satellite imagery users might have questions about the statements made above with regards to color and bit depth. It has been commonly cited in the geospatial world that 11-bit depth imagery will have far superior colors and depth into shadows and cloud cover as compared to 8-bit depth imagery. While it may well be a true statement that you can achieve a wider variety of colors in 11-bit depth imagery, many of the color differences are so subtle as to be imperceptible to the human eye. Further, when you add in the complexity of color balancing 11-bit depth on your own, the advantages of 11-bit depth as related to natural color representation are quickly diminished.

To test another commonly held feeling about 11-bit depth imagery, we offer the series of screen grabs below. In each comparison, the 11-bit image is on the left and the 8-bit depth image is on the right. We have chosen a color balance in each case that allows for maximum visibility into the shadows (the first comparison) and then into the light cloud cover (the second comparison). In our estimation, there is little to no difference in the visibility you have into areas obscured by shadows, cloud cover, haze, etc. This visual evidence is more proof that for most end users there is simply no reason to purchase 11-bit depth imagery products.

So when would one purchase 11-bit depth satellite imagery? The answer to this question relates to the loss of subtle variation in pixel values that occurs when you rescale 11-bit to 8-bit depth. While the human eye might not be able to tell the difference between two shades of aquamarine that are very close in color, there are multiple spectral analysis techniques, such as NDVI impervious surface extraction and unsupervised classification, that are highly sensitive to these subtle variations between neighboring pixel values. If one limits these subtle variations, then it also degrades the value of the selected spectral analysis techniques.

11-bit depth imagery is not only crucial for spectral analysis, it is also required when mosaicking imagery from multiple dates to create a seamless final product. To create a seamless mosaic, one of the imagery dates is color balanced and then all the other dates are matched to this color scheme. If you start with 8-bit depth imagery, there are far fewer values to ‘massage’ mathematically and you could be left with seams between the various dates where differences in the color balancing are quite obvious. A final instance where 11-bit depth imagery is essential is the detection of features whereby a unique color balance might be the only way to achieve this goal.

In sum, here is the advice eMap can offer its clients when ordering high-resolution satellite imagery as related to bit depth:

• Choose 8-bit imagery if…
• You are a novice user and only plan to use the data as a georeferenced picture.
• You need smaller file sizes.
• Choose 11-bit imagery if…
• You are completing spectral analysis on the data.
• You need to pansharpen your own data products.
• You are mosaicking across various dates/seasons/years.
• You desire to accentuate hard to locate objects in the data.

Brock Adam McCarty
Chief Operating Officer
Map Wizard