Automatic Attribute Profiles for Spectral-Spatial Classification of Hyperspectral Images

摘要

Attribute profiles integrate spectral and spatial information present in an image. The construction of attribute profiles is based on attribute filtering which requires proper threshold values. In the literature, only a few approaches are available to automatically detect the threshold values. Among them, the recently presented state-of-the-art method overcomes the prior limitations but is computationally demanding. In this paper, we present a simple and computationally efficient method to detect the threshold values automatically. The proposed method obtains the candidate threshold values directly from the tree representation of the image and selects suitable threshold values in two stages. In the first stage, we separate the larger attribute values to preserve important components and then the lower attribute values are clustered in the second stage to detect the final threshold values. Using these threshold values attribute profiles are constructed for two real hyperspectral data sets considering three different attributes. The experimental results demonstrate that the proposed method is effective and faster than the state-of-the-art method.