A nonrecursive GR algorithm to extract road networks in high-resolution images from remote sensing

• Autor(a) principal: Cardim, Guilherme Pina
• Data de Publicação: 2020
• Outros Autores: da Silva, Erivaldo Antônio [UNESP], Dias, Mauricio Araújo [UNESP], Bravo, Ignácio, Gardel, Alfredo
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1007/s12145-020-00501-5; http://hdl.handle.net/11449/202012
• topic: Algorithms; Data processing; Growing region; Image analysis

A number of studies address the development of algorithms based on the Growing Region (GR) technique adaptations for extracting road networks in images. However, these algorithms are high-computationally demanding and time-consuming while processing high-resolution images. The aim of this study is to introduce a modified version of the GR algorithm, named Nonrecursive Growing Region (NRGR), to extract road networks in high-resolution images from remote sensing. This study describes how the NRGR algorithm works to perform the extractions in a faster way. The proposed algorithm was developed taking into consideration the reduction of the data dependence between its tasks in order to allow the GR algorithm to process these tasks with the help of Graphical Processor Units (GPUs). The experiments were conducted to demonstrate the ability of the NRGR to process low or high spatial resolution images with or without the help of GPUs. Results achieved by experiments performed in this study suggest that the NRGR algorithm is less complex and faster than previous adaptations versions tested of the GR algorithm to process images. The NRGR was able to process the tested images with less than 30% of the time used by the recursive algorithm, reaching values below 10% in some cases. The NRGR algorithm can be used as software or hardware-software system’s co-design solutions to develop maps of road networks for Cartography.