Automatic threshold selection for morphological attribute profiles

Abstract

In this article, an automatized procedure for selecting informative values of the thresholds, essential for the construction of morphological attribute profiles, is proposed. To this end, connected component analysis is performed on a preliminary supervised or unsupervised classification result that does not involve contextual information. Subsequently, after extracting the relevant attributes from each of the connected components, the threshold values are found by grouping the attribute vectors using a clustering algorithm. In our experiments, we demonstrate the effect of image scaling on the selected thresholds. In addition, we show the advantage of using our automatic threshold selection approach with respect to manual selection, by both monitoring redundancy and performing a classification experiment.

Local Feature-Based Attribute Profiles for Optical Remote Sensing Image Classification

“…However, since those values might be not applicable to other data, automatic threshold selection has drawn attention from many researchers. Some interesting studies have been proposed to automatically compute attribute thresholds using fixed formulas [23], [24], supervised approaches [25], [26] as well as granulometric characteristic functions [27], [28]. Readers are referred to the mentioned papers for further details about these attribute selection strategies.…”

Automatic Attribute Profiles

“…To the authors' best knowledge, the first automatic approach aimed at decreasing the manual intervention was proposed in [36], where a vector of thresholds was derived by computing a given attribute on each object extracted by a preliminary clustering or classification computed on the original scene. The final set of thresholds was identified by clustering the threshold vector and selecting for each cluster the threshold corresponding to the minimal attribute value.…”

“…Furthermore, the classification results obtained by exploiting the proposed approach are compared against those obtained from tree strategies available in the literature and presented in [38] (hereafter Gha13), [37] (hereafter Mar13) and [36] (hereafter Mah12), taking into account their context of application (e.g., Mar13 is an approach developed to work with the standard deviation attribute, therefore is not included in the analysis when the area attribute is used). The methods are briefly described in Sec.…”

“…Although the parameter tuning is unavoidable, most of the works dealing with morphological operators for multi-level analysis do not tackle this issue, whereas the use of similar parameters, even for different case studies, seems to be the general strategy. In the literature, only few works addressing this issue can be found [36]- [39]. Since the morphological analysis is data dependent, the identification of the suitable threshold sets should be based on empirical searching.…”

Automatic Threshold Selection for Profiles of Attribute Filters Based on Granulometric Characteristic Functions

“…The common approach is based on field-knowledge of the scene, where the values are manually selected by a visual analysis of the scene under consideration [1,5]. In [6] the set of threshold was derived after a preliminary classification and clustering of the input image. In [7] the filter thresholds were chosen based on the analysis of a granulometric curve (i.e., a curve related to the size distribution of the structures in the image [8]).…”