The Effect of Partitioning of Feature Space on Specific Class Extraction Based on Bayesian Decision

Specific class extraction is becoming essential for many image processing tasks. The accuracy of specific class extraction is relative to the partitioning of feature space. The effect of partitioning of feature space on specific class extraction is studied in this paper. First, specific class extraction with different partitions of feature space is theoretically analyzed based on Bayesian decision rule. Second, an experiment on synthetic data set is performed to extract the class of interest. The experiment shows the effect of feature space partitions on the results of specific class extraction. Last, a scheme of partitioning of feature space is presented based on class separability measure and is verified in the experiment.