Feature extraction of character image using shape energy

One effort to maintain documents or records is to make changes in the form of digital images. The drawings further processing needs to be done so that the text or sentences therein can be operated as do the search, analysis, or manipulation of the contents of the text. The treatment process is known as optical character recognition (OCR) and continues to develop. OCR is generally divided into three main stages, namely preprocessing, feature extraction and classification. Feature extraction is one of the essential or fundamental processes in character recognition. The purpose of feature extraction is to obtain the characteristics of each character. The results at this stage can affect the quality of character recognition. Generally, feature extraction on character is done by a complex calculation so as to cause the necessary time computing is not a little, especially in real time recognition case. In this paper, feature extraction can be done simply proposed as an alternative, called Shape Energy. This method uses the approach of how humans are able to distinguish between characters or numbers in a simple. It results in three elements which are elasticity, curvature, and texture. The elasticity is first derivative and the curvature is second derivative of each pixel in the frame of the character, which is obtained from thinning. While the texture value is 4-direction chain-codes. This method testing has been done on some type of character by using back propagation neural network as a method on classification stage. This testing resulted in average value accuracy rate of success in identifying these characters by 90.3%.