期刊名称：International Journal of Innovative Research in Computer and Communication Engineering
印刷版ISSN：2320-9798
电子版ISSN：2320-9801
出版年度：2014
卷号：2
期号：12
出版社：S&S Publications
摘要：In this paper, we tend to describe hybrid feature extraction for offline written character recognition. Theprojected technique could be a hybrid of structural, applied math and correlation options. Within the opening, theprojected technique identifies the kind and placement of some elementary strokes within the character. The strokes to behunted for comprise horizontal, vertical, positive slant and negative slant lines–as we tend to observe that the structure ofany character are often approximated with the assistance of a mix of straightforward line strokes. The strokes are knownby correlating completely different segments of the character with the chosen elementary shapes. These normalizedcorrelation values at completely different segments of the character offer correlation options. For creating featureextraction additional strong, we tend to add within the second step sure structural/statistical options to the correlationoptions. The additional structural/statistical options are supported projections, profiles, invariant moments, endpoints andjunction points. This increased, powerful combination of options leads to a 157-variable feature vector for everycharacter, that we discover adequate enough to unambiguously represent and determine every character. Prior, writtencharacter recognition downside has not been self-addressed the means our projected hybrid feature extraction techniquedeals with it. The extracted feature vector is employed throughout the coaching section for building a support vectormachine (SVM) classifier. The trained SVM classifier is after used throughout the testing section for classifying unknowncharacters. Experiments were performed on written digit characters and uppercase alphabets taken from completelydifferent writers, with none constraint on style. The obtained results were compared with some connected existingapproaches. Attributable to the projected technique, the results obtained show higher potency concerning classifieraccuracy, memory size and coaching time as compared to those different existing approaches.
