Regioned downsample for ANN image classification: Alternate selection methods
Keith A. Greenhow, Colin G. Johnson · 2015
In an earlier paper, a novel method to pre-process image data for use in Artificial Neural-Network (ANN) classification was presented. This method requires an additional training stage prior to the main learning phase of the ANN. In this extra stage, an additional algorithm (a Selection method) is used to generate the data that is required to construct the final preprocessor. As part of the introduction of that method, it was presented with a single Selection method that was termed Saliency Heat Mapping. This paper will present a number of alternative Selection methods and compare how effective they are against a sample problem.