Personality Features Identification from Handwriting Using Convolutional Neural Networks
Sri Hastuti Fatimah, Esmeralda C. Djamal, Ridwan Ilyas, Faiza Renaldi · 2019
Many evidence suggested that one's personality can be seen from his/her handwritten scratches, using Graphology analysis. Graphology consists of two techniques, structural and symbol analysis. The structural analysis which sees handwriting structurally as a unit. While symbol analysis based on how to write each letter. With a variety of handwriting features, make identifying personalities with Graphology is not easy to do. Computer-based Graphology is done by examining the handwriting image and then giving suggested personality results based on the pattern of each feature. This research conducted using both techniques of six features. The multi-structure analysis was done to feature of margin, a space between lines, a space between words, slope, and dominant zone, while four specific letters (a', `g', `s', `t') were analyzed using the Convolutional Neural Networks (CNN) classification approach. The results showed that the accuracy of the structured approach was up to 82.5-100%, while the accuracy of the symbol approach using CNN had an accuracy of up to 98.03% of new data with 7-10 minutes in the training process.