Warning: A non-numeric value encountered in /homepages/29/d761004148/htdocs/clickandbuilds/KermitImageToolkit/wp-content/themes/Divi/functions.php on line 5560

Baddeley’s Delta metric for local contrast computation in hyperspectral imagery

by | Feb 27, 2017 |

General Information

Title: Baddeley’s Delta metric for local contrast computation in hyperspectral imagery

Journal: Progress in Artificial Intelligence

Abstract: Recent years have brought a quick decay in prize of hyperspectral imagery equipment. As a consequence, new applications have appeared, a relevant example being the analysis of agro-food materials. Such applications need to be grounded on dedicated image processing operators, which fully accomplish with (and exploit) the characteristics of hyperspectral imagery. In this regard, we study the quantitative comparison of spectra, which can be further used to produce a variety of image processing operators. Specifically, we propose the use of Baddeley’s Delta metric for the comparison of spectra. Our method has theoretical advantages over classical bandwise comparison measures, which are often inconsistent with human perception of dissimilarity between spectra. Our proposal is put to the test in the context of local contrast computation, with application to item segmentation of in-laboratory imagery.

Keywords: Hyperspectral imagery; Comparison measures; Baddeley’s Delta metric; Local contrast; Image segmentation.

Cite as: C. Lopez-Molina, D. Ayala-Martini. A. Lopez-Maestresalas, H. Bustince, Baddeley’s Delta metric for local contrast computation in hyperspectral imagery, Progress in Artificial Intelligence, January 2017, Pages 1-12.

Detailed description

Hyperspectral imaging (HSI) is a non-destructive and non-contact imaging technique that combines the principles of conventional spectroscopy and imaging techniques to provide spectral and spatial information from a sample, simultaneously. The applications of HSI to quality control of fruit and vegetables include classification/sorting, as well as identification of external and internal defects or quality estimation. In this work we aim at producing a comparison measure which can evaluate the divergences between spectra in hyperspectral imagery. Such measure shall play a crucial role in developing solid grounds for hyperspectral image comparison, as well as to abilitate the application of both classic and state-of-the-art algorithms to hyperspectral imagery.

In this work we analyze the most popular measures for spectra comparison. Specifically, we analyze the problems in band-wise spectra comparison measure, and propose to use Baddeley’s Delta Metric (BDM) instead. Moreover, we use BDM to measure the local contrast at each pixel of hyperspectral images. The validation of the local contrast operator has been carried out in the context of object segmentation of in-lab hyperspectral imagery. Specifically, in fruit and vegetable segmentation for quality analysis.


Our main conclusions relate to the improved quality and reliability of the BDM, compared to bandwise comparison measures. It is nevertheless true that such improvement is (so to say) light; also, that it comes to a certain increase in the computational cost of the local contrast computation.


More Information

Resources (in the KITT): The following resources are of interest for the study and/or use of the developments in this work:

  • Function hspectralBDM.m (package hsImagery), to compare to hyperspectra using Baddeley’s Delta Metric.
  • Function hsImage2graph.m (package hsImagery), to produce a 4- or 8-connected graph using either Baddeley’s Delta Metric or bandwise similarity measures.
  • Hyperspectral image dataset Lopez-Maestresalas 01 (LMO1) (link here).


Related works (in the KITT):

  • None, so far.

Related works (web):

  • Baddeley, A.J., “An error metric for binary images”. In: Förstner, W., Ruwiedel, S. (eds.) Robust Comput Vision: Qual Vision Algorithms, pp. 59–78. Wichmann Verlag, Karlsruhe (1992)
  • Coquin, D., Bolon, P. “Application of Baddeley’s distance to dissimilarity measurement between gray scale images”. Pattern Recognition Letters 22(14), 1483–1502 (2001)