Spatial Outlier Detection: A Graph-Based Approach

Inter-image outliers and their application to image classification

“…If μ s and σ s are the mean and the standard deviation of S, then the outlier t ∈ S can be defined as | t−μ s σ s | > 2. As can be seen in the literature [8,9], majority of methods define the candidate outliers using an approach based on Chebyshev inequality.…”

“…Spatial outlier detection methods are either based on distances [14,15,16,17], to define spatial neighborhoods, or on graph connectivity [3,4,5,6,7,8,9,10]. A comprehensive study on distance based methods is presented in [14,15].…”

“…Graph based spatial outliers [8,9,10] use the statistic S(x) = f (x) − E y∈N(x) ( f (y)) for a data record x with attribute f (x), where N(x) is the fixed cardinality set of neighbors of x and E y∈N(x) ( f (y)) is the average attribute value for the neighbors of x. In other words, the statistic S(x) denotes the difference between the attribute value of each data object x and the average attribute value of x's neighbors.…”

“…The idea of outlier detection has been explored across various fields and applications [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17] and a number of these approaches are applicable to image analysis and recognition [14,3,4,5,6,7,8,9,10]. Outliers in images are often termed as spatial outliers and can be seen to represent neighborhoods of pixels in an image that are significantly different from the rest of the image.…”

“…In the past, outliers were used in image analysis to detect changes within images [3,4,5,6,7,8,9,10,11,12,13]. However, the idea of outliers has not been explored directly for applications involving image comparison and classification.…”