Fuzzy surroundedness

A Fuzzy Rule-Based Approach to Scene Description Involving Spatial Relationships

“…SURROUND was determined using a separate neural network. Details of the approach can be found in [16,17]. We chose to calculate SURROUND as an independent measurement instead of using the conjunction of the other four relations because our findings in [16] support this independence.…”

“…Since fuzzy rule-based systems simulate human reasoning, it is natural to use them in scene description. From the spatial relationship neural networks, five spatial relation membership values (LEFT, ABOVE, RIGHT, BELOW, and SURROUND) between two individual objects were obtained by the method described in [16,17]. However, when a human being describes a scene, he/she will describe it using natural language instead of providing strict numeric values.…”

“…Spatial relations such as ABOVE, RIGHT, and others defy precise definition and seem to be best modeled by fuzzy sets [16][17][18][19][20][21][22][23]. The all-or-nothing crisp definition of spatial relations actually adds to the problem of performing reasonable scene interpretation.…”

“…Bloch [21] and Gader [23] used fuzzy dilations to create the degree of membership of two objects in the primitive relations. The second basic approach concerns making measurements from the set or the histogram of angles generated from all pairs of points, one from each object [16][17][18][19]. Matsakis and Wendling [20] recently postulated an axiomatic framework for functions that generate spatial relationships, and then constructed what they term "the histogram of forces."…”

A Study on how Humans Describe Relative Positions of Image Objects

“…The most common method of capturing human perception is as follows: To begin, subjects are presented with a small number (usually less than 100) of images, referred to as configurations, containing basic shapes. Subjects are then given a set of spatial relations (usually less than 10), and for each configuration, they are required to either answer a series of yes or no questions (i.e., whether a given relation describes the configuration (Robinson 1990)), or to rate a list of spatial relations based on how well they describe the configuration individually (Gapp 1995;Wang & Keller 1997, 1999Zhan 2002). The weaknesses in these methods are obvious.…”