An accuracy assessment determines the correctness of the classified images, which is based on object groupings. Accuracy is a measurement of the agreement between a standard that is assumed to be correct and an image classification of unknown quality. If the image classification corresponds closely with the standard, it is considered accurate.
An error (or
confusion) matrix is used as the base to calculate various metrics for accuracy assessment. The following is the list of the key metrics used:
- Kappa coefficient is the ratio of the agreement between the classifier output and reference data, and the probability that there is no chance agreement between the classified and the reference data: it is free of bias. The coefficient can be computed in two ways: among all classes and within a single class.
- Quantity disagreement measures the discrepancy in the proportions of all classes between the classified and reference data.
- Allocation disagreement assesses the aggregated misallocation of individual objects for the same level of quantity agreement.