A Brief Survey: Spatial Semi-Supervised Image Classification

Machine learning and machine classification have traditionally been done by three different methods. The first method is an unsupervised method, which takes no preclassification of any of the data in order for the data to be grouped together for analysis afterwards. This method then requires a domain specialist to examine each group after the results have been run to label the resulting groups of data. A second method is a supervised approach. In this method, a large number of data points must be evaluated prior to the grouping of the data, in order to create the groups. This method requires a considerable amount of domain specialist time in order to classify the data in order to provide adequate results. The third method combines these two approaches to begin with a set of domain specialist specified data, (although considerably less data than in the supervised method) and then using a set of random data points that have not been specified. Within this method there have been a considerable number of approaches within the traditional non-spatial data mining methods. What makes the spatial variety of semi-supervised learning and classification different is that the use of traditional semi-supervised learning makes the assumption that each data point, or observation is independent of the surrounding observations [5]. This assumption however does not hold in the spatial realm of data [6]. This must then be accounted for in the evaluation of the semisupervised classification methods. In comparison to this paper, a survey of traditional semi-supervised learning can be found from [7]. This survey is organized as follows. In the next section we will give a brief overview of spatial classification methods and the limitations of each method. In Section 3, we will examine the principles of spatial semi-supervised learning. Section 4 will provide an overview of what areas have been explored in the spatial semi-supervised method. Section 5 will discuss some possible areas of future work followed by section 6 concluding the work.