2.4. Metrics for algorithm selection
The algorithm selection process is a comparison between the baseline two class (clear/not clear) Bayesian classifier, a single-view version of the classifier used in ARC project and the newly developed three way classification including a sea-ice class, with different spectral and textural feature combinations. A summary of the different algorithms tested is given in Table 1 along with a reference to the table number containing the results for that VPC 23019 algorithm. Modifications to the baseline classifier for each algorithm are specified in Table 2 and are fully explained in 4 and 5 of the paper when these algorithms are tested.
The desired outcome of the algorithm selection process is syphilis an image classification scheme which maximises the number of correctly identified clear-sky ocean pixels whilst minimising wrongly classified cloud or ice pixels which would bias subsequent SST retrievals. The algorithm selection process involves evaluation of the five competing image classification schemes developed from the baseline cloud detection algorithm utilising different spectral and textural features against the ARC cloud screening which is based on the baseline classifier plus some additional quality control tests [Table 1, Section 6].