A New Aerial Image Segmentation Approach with Statistical Multimodal Markov Fields

Aerial images, captured by drones, satellites, or aircraft, are omnipresent in diverse fields, from mapping and surveillance to precision agriculture. The efficacy of image analysis in these domains hinges on the quality of segmentation, and the precise delineation of objects and regions of interest. In this context, leveraging Markov fields for aerial image segmentation emerges as a promising avenue. The segmentation of aerial images presents a formidable challenge due to the variability in capture conditions, lighting, vegetation, and environmental factors. To meet this challenge, the work proposes an innovative method harnessing the power of Markov fields by integrating a multimodal energy function. This energy function amalgamates key attributes, including color difference measured by the CIEDE2000 metric, texture features, and detected edge information. The CIEDE2000 metric, derived from the CIELab color space, is renowned for its ability to measure color difference more consistently with human perception than conventional metrics. By incorporating this metric into the energy function, the approach enhances sensitivity to subtle color variations crucial for aerial image segmentation. Texture, a vital attribute characterizing regions in aerial images, offers crucial insights into terrain or objects. The method incorporates texture features to refine the separation of homogeneous regions. Contours, playing a fundamental role in segmentation, are identified using an edge detector to pinpoint boundaries between regions of interest. This information is integrated into the energy function, elevating contour consistency and segmentation accuracy. This article comprehensively presents the methodological approach, the conducted experiments, obtained results, and a thorough discussion of the method's advantages and limitations.

@article{Bouchti2024,
  title     = {A New Aerial Image Segmentation Approach with Statistical Multimodal Markov Fields},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {3},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Jamal Bouchti and Ahmed Bendahmane and Adel Asselman},
  doi       = {10.14569/IJACSA.2024.01503106},
  url       = {https://doi.org/10.14569/IJACSA.2024.01503106}
}