TEmoDec: Emotion Detection from Handwritten Text using Agglomerative Clustering

Handwriting analysis is the practice of understanding a person better by examining their handwriting. The traditional approach involves the examination of several parameters like Pen-Pressure, Slant, Baseline, Zone, Margin, and Size of the handwritten text, by an expert. These parameters are indicative of the mental state of the person and can be used to judge the honesty, stress, and depression levels of the person. The disadvantage of this practice is that it is time-consuming and accuracies vary according to the skills of the examiner. In this paper, we propose a novel method based on the Agglomerative Hierarchical Clustering technique, that is able to identify the emotional state of the person by looking at the image of the handwritten text. Thus we are able to achieve reliable accuracies without the need for large annotated datasets by using unsupervised learning. After preprocessing, the image pixels are clustered, based on a threshold value of intra-cluster distances. All pixels with distance, from the centroid of the cluster, lower than the threshold value belong to that cluster. Each cluster corresponds to one of the predefined emotions. We test our model to predict 5 emotions, namely, Anger, Sadness, Depression, Happiness and Excitement. However, the proposed method can be used for more emotions as necessary by changing the threshold distance value. We achieve an accuracy above 75% for each of these emotions. Our work may potentially be used in mental health diagnosis, in the hiring process by the industry as well as in criminal investigation.