Pre-Processing and Feature Extraction Technique for Hand-drawn Finite Automata Recognition

Recognition of hand-drawn images is an easy task for human beings, but a challenging and cumbersome task for the computer to do automatically due to many factors including handwriting variations. This paper focuses on image pre-processing and feature extraction and proposes an effective offline technique to process handdrawn finite automata (FAs). Pre-processing techniques are the first approach in a recognition system. We applied a number of pre-processing techniques to 20 handdrawn FA images to see how feasible it is to extract the primitives of the input FA images. The output of the pre-processing stage was used as an input for the feature extraction stage. The histogram of oriented gradients (HOG) technique was used to extract the features in an FA diagram. A HOG descriptor takes the visual components of an input FA and describes them by the edge directions. The result of our experiment shows that using a HOG algorithm reduces the dimensionality of the feature space without disturbing the performance of the classifier, thus maintaining high efficiency with little computational complexity. Also, our experiment result shows that our system produced good results that can be used in classification.