Browsing by Author "Koc,S.G."

Now showing 1 - 3 of 3

Citation Count: 4
Attribute profiles without thresholds
(Institute of Electrical and Electronics Engineers Inc., 2018) Aptoula,E.; Koc,S.G.
Morphological attribute profiles are among the most prominent spatial-spectral pixel description methods. They are efficient, highly flexible multiscale tools that operate at the connected component level of images. One of their few yet significant drawbacks is their need for a predefined threshold set. As such there have been multiple attempts for computing thresholds with minimal or no supervision with various levels of success. In this paper, a radically different approach is taken and a new way is presented, circumventing the need for thresholds while harnessing the descriptive power of the hierarchical tree representation underlying the attribute profiles. The introduced approach is validated with two datasets and two attributes, where it exhibits either comparable or superior performance to manual and automatic threshold based attribute profiles. © 2018 IEEE.
Citation Count: 3
A comparative noise robustness study of tree representations for attribute profile construction;
(Institute of Electrical and Electronics Engineers Inc., 2017) Koc,S.G.; Aptoula,E.; Bosilj,P.; Damodaran,B.B.; Mura,M.D.; Lefevre,S.
Morphological attribute profiles are among the most prominent spatial-spectral pixel description tools. They can be calculated efficiently from tree based representations of an image. Although mostly implemented with inclusion trees (i.e. component trees and tree of shapes), attribute profiles have been recently adapted to partitioning trees, and specifically α- and ω-trees. Partitioning trees constitute a more flexible option especially when dealing with multivariate data. This work explores the noise robustness of the aforementioned major tree types in terms of pixel classification performance of the resulting attribute profiles, and presents our preliminary findings that support the use of partitioning trees as a basis for attribute profile construction. © 2017 IEEE.
Citation Count: 4
Hyperspectral image classification with convolutional networks trained with self-dual attribute profiles;
(Institute of Electrical and Electronics Engineers Inc., 2017) Koc,S.G.; Aptoula,E.
Attribute profiles are widely regarded among the most prominent spectral-spatial pixel description methods, providing high performance at a low computational cost. Following their success with computer vision applications, deep learning methods on the other hand are also being rapidly deployed and adapted into the remote sensing image analysis domain, where they already provide competitive description performances. The combination of attribute profiles with convolutional neural networks has recently taken place, showing that these powerful approaches can collaborate. In this paper we explore that direction one step further, by first feeding a convolutional neural network self-dual attribute profiles stacked as a tensor, and then by harvesting the ultimate layer's features for a supervised classification. Our preliminary experiments indicate that this approach leads to a performance improvement. © 2017 IEEE.