|
Non-linear approaches based on the maximum distance — a pseudo morphology and PCA approximation for color, multispectral and hyperspectral data/image analysis| old_uid | 16127 |
|---|
| title | Non-linear approaches based on the maximum distance — a pseudo morphology and PCA approximation for color, multispectral and hyperspectral data/image analysis |
|---|
| start_date | 2018/07/06 |
|---|
| schedule | 14h-15h |
|---|
| online | no |
|---|
| location_info | Auditorium J. Herbrand |
|---|
| summary | n the context of complexity assessment and texture image characterization for color natural and fractal images, we start by introducing a probabilistic pseudo-morphology based on the Chebyshev inequality. We present our experimental results on complexity assessment by fractal measures, as well as image segmentation based on tailored texture features. Further on, we propose a maximum-distance-based pseudo-morphology and show that extending the existing morphological approaches for texture characterization from grayscale to color and to multispectral images in a straight-forward way is not able to make full usage of the spectral information acquired by the sensors. Then the same maximum distance is used for developing a non-optimal geometric approximation of the principal component analysis (PCA). We validate this approach on synthetic 2D data and show its usefulness for the visualization and analysis of hyperspectral images. We conclude on how outliers of a multivariate data set can provide information compared to the majority, as in classical statistics. |
|---|
| responsibles | <not specified> |
|---|
| |
|