Artificial neural network-based lot number recognition for cadastral map

Abstract

This paper discusses the implementation of an artificial neural network in detecting lot numbers in a cadastral map. Specifically, the image processing techniques used in the study are binarization, connected component labeling, and image resizing. A feed-forward with backpropagation artificial neural network was then implemented in the training and learning activities of the system. Fifty different maps with an average of 31 lot numbers at size 832 x 768 pixels were tested and it yielded an average of 90% detection after the 50th training. A 5808 x 4256 pixels map with 237 lot numbers was tested fifty times and gave an average of 84.78% detection rate. The satisfactory percentage of learning and detection based from the different test scenarios have supported that the implementation of the methods and algorithms is sufficient and accurate. © 2013 Springer-Verlag GmbH.