Comparative Study Between Object-Based Image Analysis Using Mean Shift and Multiresolution Segmentation Algorithms for Green Open Space Identification: Case Study in Sleman Regency

Abstract

We elaborate on the critical importance of green open spaces in the context of urban sustainability. Public health is inextricably linked to the green open spaces in urban regions, as green spaces contribute to the ecological balance. In Indonesia, Green Open Space (GOS) is guided by Law No. 26 of 2007, which designates at least 30% of urban areas as GOS. However, tracking GOS is challenging due to urbanization and a lack of ground surveys. Remote sensing techniques and Object-Based Image Analysis (OBIA) may help solve this problem. We used high-resolution SPOT-7 imagery to determine GOS in Sleman Regency, Indonesia. The study also compares two OBIA segmentation algorithms, Mean Shift and Multiresolution Segmentation. The classification included city parks, rice fields, gardens, buildings, water bodies, and areas labeled as unclassified. Mean Shift segmentation, combined with SVM, achieved an accuracy of 81.25% and a Kappa index of 0.739. On the other hand, Multiresolution Segmentation with Nearest Neighbour attained 91.25% accuracy and a Kappa index of 0.878. Although these results indicate the superiority of Multiresolution Segmentation over Mean Shift in heterogeneous urban settings, there is a potential for the latter in fine-scale feature detection.