Image-based Estimation of Physical Correlates of the Visual Appearance of Snow

Abstract

The most common representation of snow is to describe it as a white and cold powder material that is usually found in winter or specific areas of the world. It can also be associated with the Christmas festivities, winter sports and leisure activities such as alpine or cross-country skiing and more recently with global warming issues or the seeking of water on other planets. As human being, the appearance of snow can be distinguished thanks to several visible interactions on the surface of snow but is also influenced by phenomena occurring under its surface and connected to its microstructure.

Capturing visual appearance features of snow is a challenging task as the natural and unstable characteristics of the material often require operating at the limits of the sensor capacities. This thesis aims to utilize image-based methods to acquire various snow correlatives related to visual appearance, gather them, and analyse to try to find links with a potential classification of snow.

A first part is dedicated to the investigation of a reflectance model of snow by using hyperspectral cameras to find back values of the snow grain size and the snow grain shape. These estimates can be used to establish a classification of the type of snow. We performed acquisitions in a laboratory with snow samples, and we monitored the evolution of melting snow by obtaining hyperspectral images. From these images, we derived an effective parameter to qualify the contribution of both snow grain size and snow grain shape, although we could not obtain a precise and distinct measurements of these two parameters due to a lack of ground truth data.

Secondly, the sparkle of snow is measured from digital images acquired insitu over two winters. Datasets of snow images were established by performing outdoor acquisitions with a DSLR camera. A state-of-the-art algorithm originally designed to measure sparkle was adapted to the case of snow. With a statistical analysis of the results, an attempt at finding a connection between sparkle and categories of snow is made. A classification seems possible, but further investigations with an expert and precise labelling should be operated to confirm this theory.

Finally, an inversion method is designed to obtain estimates of absorption and scattering properties of highly diffuse materials with a single reflectance measurement. After being tested and validated on dairy products, an in-situ campaign was operated during a winter by taking the measuring device outside. The results obtained from this study confirm the absorption and scattering properties of snow, while opening new perspectives for the virtual rendering of this material.

The work achieved gives contributions to various research areas, all connected with imaging methods. In addition, we started to link our results from different correlates with intrinsic parameters of snow such as grain size and grain shape, thus paving the way for extension of research in this field.