Hilda Deborah

A Comprehensive Evaluation of Spectral Distance Functions and Metrics for Hyperspectral Image Processing

Distance functions are at the core of important data analysis and processing tools, e.g., PCA, classification, vector median filter, and mathematical morphology. Despite its key role, a distance function is often used without careful consideration of its underlying assumptions and mathematical construction. With the objective of identifying a suitable distance function for hyperspectral images so as to maintain the accuracy of hyperspectral image processing results, we compare existing distance functions and define a suitable set of selection criteria. Bearing in mind that the selection of distance functions is highly related to the actual definition of the spectrum, we also classify the existing distance functions based on how they inherently define a spectrum. Theoretical constraints and behavior, as well as numerical tests are proposed for the evaluation of distance functions. With regards to the evaluation criteria, Euclidean distance of cumulative spectrum (ECS) was found to be the most suitable distance function.

Pigment Mapping of the Scream (1893) Based on Hyperspectral Imaging

Hyperspectral imaging is a promising non-invasive method for applications in conservation of painting. With its ability to capture both spatial and spectral information which relates to physical characteristics of materials, the identification of pigments and its spatial distribution across the painting is now possible. In this work, The Scream (1893) by Edvard Munch is acquired using a hyperspectral scanner and the pigment mapping of its constituent pigments are carried out. Two spectral image classification methods, i.e. Spectral Angle Mapper (SAM) and Spectral Correlation Mapper (SCM), and a fully constrained spectral unmixing algorithm combined with linear mixing model are employed for the pigment mapping of the painting.

Spectral Ordering Assessment Using Spectral Median Filters

Distance-based mathematical morphology offers a promising opportunity to develop a metrological spectral image processing framework. Within this objective, a suitable spectral ordering relation is required and it must be validated by metrological means, e.g. accuracy, bias, uncertainty, etc. In this work we address the questions of suitable ordering relation and its uncertainty for the specific case of hyperspectral images. Median filter is shown to be a suitable tool for the assessment of spectral ordering uncertainty. Several spectral ordering relations are provided and the performances of spectral median filters based on the aforementioned ordering relations are compared.

Toward a full-band texture features for spectral images

Facing the increasing number of multi and hyperspectral image acquisitions, in particular for medical and industrial applications, we need accurate features to analyse and assess the content complexity in a metrological way. In this paper, we explore an original way to compute texture features for spectral images in a full-band and vector process. To do it, we developed a dedicated approach for Mathematical Morphology using distance function. Thanks to this, we extend the classical mathematical morphology to spectral images. We show in this paper the scientific construction and preliminary results.

On the Quality Evaluation of Spectral Image Processing Algorithms

The main contribution of this work lies in the context of the quality assessment of spectral image processing operators or algorithms. Several means to define quality measurements for low-level spectral image processing tools are explored, i.e. Spectral distance and spectral image filtering. The quality assessment starts from simple theoretical validation on simulated spectra to reduced and full reference quality assessment on real spectral data. Spectral image of a pigment patch with different shades and a simple model of spectral noise are proposed to be used in reduced and full reference quality assessment. Finally, given an application purpose, this quality assessment protocol will aid reader in selecting a correct spectral distance which will be the core of other distance-based spectral image processing tools.

Hyperspectral crack detection in paintings

Several approaches to the crack detection of paintings are available for grayscale and color images, and recently also for spectral images. However, the approaches that are used for the multivariate data are either using a marginal approach or requiring a data reduction which enable the use of grayscale operators. In this study, the crack detection task is addressed with a spectral processing expressed in a fullband and vector approach. By using distance functions in the ordering relations and crack detection method, the metrological constraints required by such important cultural heritage objects are respected. The performances of the crack detection methods are assessed with artificial images which combine real spectral images of known properties and simple probabilistic crack model, and also with images from cracked paintings.

Spectral Impulse Noise Model for Spectral Image Processing

The performance of an image processing algorithm can be assessed through its resulting images. However, in order to do so, both ground truth image and noisy target image with known properties are typically required. In the context of hyperspectral image processing, another constraint is introduced, i.e. apart from its mathematical properties, an artificial signal, noise, or variations should be physically correct. Deciding to work in an intermediate level, between real spectral images and mathematical model of noise, we develop an approach for obtaining suitable spectral impulse signals. The model is followed by construction of target images corrupted by impulse signals and these images will later on be used to evaluate the performance of a filtering algorithm.