Statistical Analysis Skills
The statistical analysis of features extracted from complex image objects permits robust characterisation of the objects. For example a skin lesion may be characterised as “benign” or “melanoma” or an extruded product classified as “within normal limits” or “outside normal limits”. On other occasions the goal may be to create an abnormality index or quality index – a number related to the particular property of the image being analysed, for example the probability of the sample being abnormal. The image analysis group contains several statisticians with extensive experience in applied statistics.
Various statistical analysis technologies are available, each with particular strengths and suitabilities to certain classes of problems. These include:
- linear discriminant analysis and its various generalisations (quadratic-, regularised-, mixture- discriminant analysis)
- statistical modelling: linear models, generalized linear models, robust and resistant regression, non-linear regression, additive models
- neural networks
- classification and regression trees (CART), C4.5
- spatial statistics and characterisation of spatial association
- boolean models
- nonparametric regression and smoothing
- parameter estimation and over-fitting issues.
Many of these analysis techniques are included in the statistical package “R” on which our image analysis segmentation and feature extraction software solutions can be based. This enables a seemless integration of image processing and analysis, feature extraction, and statistical analysis within a common framework and facilitates the delivery of a complete end-to-end application to the client.
Statistical analysis was an important part of the following projects:
- Melanoma Diagnosis
- Roadcrack Detection
- Food Quality