Hierarchical Scale-space Representational Measure for Estimating Land Cover
Current Topics on Mathematics and Computer Science Vol. 2,
12 June 2021
The Minimum Mapping Unit (MMU) for an object-oriented image analysis operation are shape-filling curves such as planar lines and rectilinear segments. The space-filling curves do not change the feature object representation when scale is varied, thus representing spatial and aspatial features with finer or coarser granularity. The increased collinearity can be explained by arrangement of topological objects in the aggregated feature space (neighbors/objects), thus producing image areal objects. Rather than performing one individual operation on the imagery objects by scanline rows, we can compute on custom built algorithms applied to the distinguishing objects. This operation outputs super-objects that are classified by texture and mathematical relationships, thereby leveraging the multi-scale object-oriented analysis procedures. For information retrieval, a continuous hierarchical scale space filtering operation is adapted for segmentation purposes. In fact, MMU variations will produce instances of image objects that preserve the spatial scale at a particular optimizing parameter. This article lays emphasis on object-oriented analysis and accompanying fuzzy inference analysis of the imagery scene. By denoting image analysis procedures based on image objects at the characteristic scale, one can delineate imagery semantics at the low and how-level spatial context.
Such a method becomes feasible with object oriented scale space hierarchical theory, with varying intra and inter scale parameters. While using image objects to calculate multi-variate statistics (Entropy measure, heterogeneity measure, local mean vs local variance measure, and mean vs. covariance measure), fuzzy modeling of mixed pixels is used to extract reliability without incorporating edges. Homogeneous areas of mixed pixels will be resolved by the Region Labeling Operator using in-class variance measures. Between class variances can be used to measure the distance of scale intervals that can be resolved by the scale object. This produces a hierarchical network that further delineates the final object features. The Scale Operator (SO) is defined to be the varying optimizer selection in the Region Growing and Region Merging procedures. While conducting region abstraction process, individual objects having similar sub-class variance, sub-class texture characteristics, will be fused to create a segmented super object. With the resulting increase in heterogeneity, the Scale Operator diffuses the super-objects and so more objects are merged and created within the class intervals.
- Hierarchical scale space
- feature extraction
- fuzzy inference system
- landsat- 8
- fuzzy knowledge rules
- object oriented image analysis