Shadow detection using object area-based and morphological filtering for very high-resolution satellite imagery of urban areas

Autor(es)

Azevedo, Samara ^[1] ; Silva, Erivaldo ^[2] ; Colnago, Marilaine ^[3] ; Negri, Rogerio ^[4] ; Casaca, Wallace ^[3]

Afiliação do(s) autor(es)

^[1] Univ Fed Itajuba, Nat Resources Inst, Geomat, Itajuba, MG - Brazil

^[2] Sao Paulo State Univ, Dept Cartog, Sao Paulo - Brazil

^[3] Sao Paulo State Univ, Dept Energy Engn, Sao Paulo - Brazil

^[4] Sao Paulo State Univ, Dept Environm Engn, Sao Jose Dos Campos - Brazil

• Tipo de documento: Artigo Científico
• Fonte: JOURNAL OF APPLIED REMOTE SENSING; v. 13, n. 3 AUG 9 2019.
• Citações Web of Science: 0

Resumo

The presence of shadows in remote sensing images leads to misinterpretation of objects and a wrong discrimination of the targets of interest, therefore, limiting the use of several imaging applications. An automatic area-based approach for shadow detection is proposed, which combines spatial and spectral features into a unified and flexible approach. Potential shadow-pixels candidates are identified using morphological-based operators, in particular, black-top-hat transformations as well as area injunction strategies as computed by the well-established normalized saturation-value difference index. The obtained output is a shadow mask, refined in the last step of our method in order to reduce misclassified pixels. Experiments over a large dataset formed by more than 200 scenes of very high-resolution images covering the metropolitan urban area of Sao Paulo city are performed, where the images are collected from the WorldView-2 (WV-2) and Pleiades-1B (PL-1B) sensors. As verified by an extensive battery of tests, the proposed method provides a good level of discrimination between shadow and non-shadow pixels, with an overall accuracy up to 94.2%, for WV-2, and 90.84%, for PL-1B. Comparative results also attested that the designed approach is very competitive against representative state-of-the-art methods and it can be used for further shadow removal-dependent applications. (C) 2019 Society of Photo-Optical Instrumentation Engineers (SPIE) (AU)

• Processo FAPESP: 13/25257-4 - Detecção automática de sombras e remoção dos seus efeitos em imagens digitais de alta resolução espacial
• Beneficiário: Samara Calçado de Azevedo
• Modalidade de apoio: Bolsas no Brasil - Doutorado