Computer method for 3-D images classification
Extracting salient and robust features from images may often be affected by scaling, normalization, and orientation problems; generality and performance are often compromised. Our analysis method for automatic characterization and classification on any n-D images is able to capture global properties by computing features from image isosurfaces offering suitable alternative to standard methods.
Extracting salient and robust features from images may often be affected by scaling, normalization, and orientation problems; generality and performance are often compromised. These problems are particularly relevant for low resolution and low contrast digital 3-D images. Although this analysis has been developed and optimized for PET images of the brain, it is a general geometrical method for automatic characterization, ranking and classification, similar to human visual analysis. It can be applied to any set of homogeneous data that represents a population of objects described by n-D scalar matrices. ELBA, its lightweight and fast algorithm, is able to capture global relevant properties of each image by computing features from image isosurfaces offering suitable alternative to comparing intensity-based methods. Features provided are sufficiently general to be applied to any kind of image summarizing its content.
- Clinical brain amyloid PET evaluation targeted to drug trial monitoring;
- Supernovae detection in astronomical images;
- Background filtering and object recognition;
- Characterization of images time series;
- Data mining;
- Night vision enhancement;
- Flaws classification and monitoring.
- Provides features, stable and easy to obtain, computable on any n-D matrix;
- Perceptual approach closer to visual assessment;
- Provides complementary info combinable with existing methods;
- Avoid intensity normalization;
- Extremely generalizable and flexible;
- Not strictly requires image registration.