–AI and learning model improvements

–image segmentation accuracy and parameter learning

–complimentary segmentation methods including real-time micro-segmentations

–re-segmentations of interesting regions in multiple image and color spaces

–developing better distance functions and learning distance function parameters

–learning complimentary distance functions,

–learning feature matching thresholds for scoring

–learning the best features for a specific classifier or set of classifiers

–automating the code generation and parameterization of classifiers

–and learning multi-level classifier group structures instead of relying on a single classifier

–Multiple segmentation API: most critical features include segmentation learning to identify the best parameters and region sizes, including targeted segmentations to support a range of image resolutions.

–Dynamic segmentation API: an interactive tool interface providing several choices of genome segmentations around a feature of interest, like a segmentation display and picker.

–Image space permutation API: allow for the selection of multiple genome cross-compare permutations in various image spaces such as reference:[raw, sharp, retinex, histeq, blur] target:[raw, sharp, retinex, histeq, blur] to find the optimal correspondence. This will be a high-level API which calls the MCC functions several times with various permutation ordering and parameterization and collects the results into a CSV.

–Color space permutation API: provide additional API support to agents for the selection of multiple color space compare permutations: reference[color component, color_space, leveled-space]: target[color component, color_space, leveled-space]. This will be a high-level API which calls the MCC functions several times with a set of permutation orderings and parameterizations, and collects the results into a CSV.

–Expanded volume processing spaces: provide parameter options for additional sharp, retinex, blur, and histeq volumes over a sliding range of strength, such as raw → soft_sharpen → medium_sharpen → heavy_sharpen.

–Chain metric qualification API: start from selected reliable metrics as qualifier metrics, creating chains of candidate dependent metrics, similar to the boosting methods in ADABOOST [1].

–Expanded autolearning hull families: provide additional autolearning hull defaults to cover image permutations, metric spaces, and invariance criteria. Hull selection will be parameterized into the MCC functions, with parameterized hull family code generation.

–Expanded scoring API: provide API support in MCC functions for a family of autolearning hull computations tuned for each type of metric, taking better account of the metric range, using separate algorithms for volume hulls, color hulls, Haralick hulls, SDMX hulls. Will include a default Softmax classifier.

–Color autolearning hull filtering: filter out unwanted colors due to poor segmentations by reducing the hull to the most popular colors.

–Qualifier metric selection overrides API: a learning method to pair and weight dependent metrics against trusted metrics.

–Low-bit resolution VDNA histogram API tool: for grouping and plotting metrics groups to visualize gross metric distance for selected metrics in an image or a group of images.

–Alignment spaces specification API: to provide additional MCC parameters for 3D volume comparison alignment of target to reference volume, using 3D centroids as the alignment point. Besides centroid alignment, other alignments will be added such as sliding volume alignment (similar to the sliding color metrics). Additional 2D alignment spaces will be added to align target to reference values within color leveling alignment spaces and other alignment markers, such as min, max, ave, and centroid. (NOTE: centroid marker alignment is currently supported for MCC volume related functions via a global convenience function). Add a new alignment space model for squinting using a common contrast remapping method.