Testing for sources of errors in quantitative image analysis

Because image analysis is subject to errors, this chapter advocates for the systematic performance of quality controls of the results. Errors can be reduced to a minimum, but to do so it is necessary to identify their cause. Since an image is the result of a long series of sampling and operations, many factors can introduce error. Each parameter can be tested by slightly varying that parameter while maintaining the other constant and monitoring their impact on the final measurement. Preparation (or acquisition) errors are easy to test but require performing some empirical trials. Integration errors (or sampling errors) can be substantially limited if basic rules are respected (Table 6). Analysis errors (or errors due to processing the images) require that the transformation applied to the image is fully understood by the operator. There is a great risk of bias if a processing algorithm is used as a magic black box. Finally, efforts should also be made to try to validate the results with external methods and correctly educate geoscientists in image processing and analysis.