Benchmarking of Remote Sensing Segmentation Methods

Abstract

We present the enrichment of the Prague Texture Segmentation Data-Generator and Benchmark (PTSDB) to include the assessment of the remote sensing (RS) image segmenters. The PTSDB tool is a Web-based (http://mosaic.utia.cas.cz) service designed for real-time performance evaluation, mutual comparison, and ranking of various supervised or unsupervised static or dynamic image segmenters. PTSDB supports rapid verification and development of new segmentation approaches. The RS datasets contain ten spectral Advanced Land Imager (ALI) satellite images, their RGB subsets, and very-high-resolution GeoEye RGB images, with optional additive-noise-resistance checking. Alternative setting options allow us to also test scale, rotation, or illumination invariance. The meaningfulness of the newly proposed dataset is demonstrated by testing and comparing several RS segmentation algorithms, and showing that the benchmark figures provide a solid framework for the fair and critical comparison among different techniques.

Operational Large-Scale Segmentation of Imagery Based on Iterative Elimination

“…It is difficult to compare segmentation algorithms because comparison involves multiple criteria and depends on the specific application and scale [51]. However, we have demonstrated that the algorithm proposed in this paper compares favourably in most assessment metrics with the commonly-used mean-shift [42] and multi-resolution [28] segmentation approaches when applied to a typical rural and forest landscape in New Zealand.…”

“…To compare the performance of the algorithm detailed in this paper to those of others, four alternatives were identified. The algorithms used for the comparison were the mean-shift algorithm [42] implemented within the Orfeo toolbox [47], as it is widely cited (e.g., [48][49][50]) as an approach that produced good results on a wide variety of Earth Observation (EO) data, the eCognition multi-resolution segmentation algorithm [28], as the algorithm most commonly used within the literature (e.g., [1,2,51]), and the Quickshift algorithm of Vedaldi and Soatto [52] and the algorithm of Felzenszwalb and Huttenlocher [53], implemented within the scikit-image library and interfaced within the RSGISLib library [54], as examples of more recent approaches from the computer vision community applicable to EO data. A SPOT-5 scene (a subset of which is shown in Figure 10A), which represents a range of land covers and uses, was used for the experiment.…”

A Class-Independent Texture-Separation Method Based on a Pixel-Wise Binary Classification

“…In this subsection, the presented technique will be applied to mosaics comprising regions made of remote sensing images from the GeoEye RGB images. The images are also available in the Prague Texture Segmentation Datagenerator and Benchmark [ 99 ]. It should be noted however, that as the mosaic structure presented in [ 99 ] the mosaics created for this paper only approximately correspond to satellite scenes.…”

“…To quantify the results, 10,000 mosaic images were generated using remote sensing images from the Prague Texture Segmentation Datagenerator and Benchmark [ 99 ]. The area under the mean precision–recall curve, the maximum mean F-measure, considering all images, and the maximum mean Pratt Figure of Merit, also considering all images, are presented in Table 5 .…”

Species-Level Vegetation Mapping in a Himalayan Treeline Ecotone Using Unmanned Aerial System (UAS) Imagery

“…Before the detailed classification of vegetated objects, a spectral difference segmentation (SDS) was performed only on the objects classified as green vegetation in level 1 classification. SDS merges neighboring objects if the difference in their spectral mean is below a given threshold (i.e., the objects are spectrally similar) [51]. Again, the optimal SDS parameters were determined by visually comparing results of several iterations.…”