Optical character recognition (OCR) on electrical specification plates

Abstract

In this paper we have developed an Optical Character Recognition (OCR) system to be used on electrical specification plates. The project is given by the Stavanger based company Verico AS(Verico). Verico performs large scale of Asset Documentation, where photo documentation is one of the main tools. Today the data collection done from the photo documentation are performed manually, which is a monotonous and time consuming process. Our system is developed to streamline their data collection process, by automatically reading specification data from images captured of electrical specification plates. The system contains two main sections, a preprocessing part and a character recognition part. The preprocessing part performs several image processing operations including background segmentation, character segmentation and several operations to prepare the image for classification. Background segmentation is performed using Otsu`s thresholding method. For character segmentation we use vertical histogram analysis. We also present a modified version of vertical histogram analysis for splitting of connected characters. The character recognition consists of two main parts, feature extraction and classification. We use direction extraction as our feature extraction method. This method looks at direction transitions in thinned version of the character one wish to classify. This method is originally presented as a method for recognition of handwritten characters, in our work we present the accuracy we obtained using this method against electrical specification plates. Classification is done using k-nearest-neighbor method.