Attribute Profiles from Partitioning Trees

Abstract

International audienc

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

“…multi-and hyperspectral data) is not straightforward. That is why the authors in [21] have recently investigated and proposed to use the partition trees such as α-tree and ω-tree to compute the α-APs, ω-APs, respectively. These profiles have been proved to provide fair performance compared to the standard APs.…”

Feature Profiles from Attribute Filtering for Classification of Remote Sensing Images

“…original APs), other implementations have been proposed using the tree of shapes [4], [5] (i.e. self-dual APs) or the partition trees (such as αtree and ω-tree to produce α-profiles, ω-profiles, respectively) [6]. In this work, we focus on the standard APs proposed in [2] using min-tree and max-tree.…”

“…For example, as the standard APs [2] were generated based on a mintree and a max-tree, some of these studies have focused on advanced implementations using other tree structures such as the tree of shapes [4], [5] or the partition trees (binary partition tree, α-tree, ω-tree, etc.) [6]. Other works have proposed to extract the histogram [7] or the first-order statistical features [8], [9] of the local patch around each AP sample for better characterization of textures within high resolution images.…”

Classification of Remote Sensing Images Using Attribute Profiles and Feature Profiles from Different Trees: A Comparative Study

“…Instead of calculating the APs based on both max-tree and min-tree image representation as in the original work [1], other implementations have been proposed using the inclusion tree (tree of shapes) [4] (i.e. selfdual APs or sd-APs) as well as the partition trees such as α-tree and ω-tree to produce α-APs, ω-APs, respectively [5]. The advantage of these trees is that their self-dual property enables the attribute filtering to simultaneously access and model both dark and bright regions from the image.…”

“…The dimension of their APs is therefore reduced by half. Moreover, partition trees offer the possibility to work on multivariate images only using a single tree, which is not trivial by exploiting component or inclusion trees [5]. In this work, those different tree structures are also adopted for the implementation of FPs.…”