Strategies for Learning the Contrast Enhancement Experts

To train a knowledge-based contrast enhancement component, the mapping between the input gray-scale feature vectors discussed in the previous section and the target method indicated by the quantitative measure of segmentation from Table 11.2 is learnt. This section gives an overview of two strategies to learn the target enhancement mapping:

Double network mapping (DNM): This strategy adopts a divide-and-conqueror paradigm. It attempts to decompose a single mapping into two simpler mappings. The first mapping to be learnt between the features from ROI and the three quantitative measure of enhancement performance proposed in section 11.3.1. A second process learns the mapping of the quantitative measure of enhancement with quantitative measure of segmentation. On testing this strategy will predict a measure of segmentation for each contrast enhancement method, and the actual contrast enhancement method is identified as the one maximizing the segmentation performance.

Breast profile mapping (BPM): This strategy differs in that the solution adopts a classification-based approach and aims to learn the mapping of feature set Fbp extracted from the complete breast image with a label of the target enhancement. On test, a single contrast enhancement method is predicted. Each strategy is described in detail in the following sections.

0 0

Post a comment