Symbolic Approach for Word-Level Script Classification in Video Frames

In recent years, addiction towards internet and digital world has made difficult for people to understand multilingual scripts in various circumstances. In this work, we proposed a model for classification of South Indian multilingual word script extracted from video frames namely, Kannada, Tamil, Telugu, Malayalam and English. Firstly, we extracted Local Binary Pattern (LBP), Histogram of Oriented Gradients (HoG) and Gradient Local Auto-Correlation (GLAC) features for each multilingual word script. The multilingual word script consists of five classes and each class of images are clustered by implementing k-means clustering technique. Further, we proposed symbolic representation to capture intra-class variations between each clusters and symbolic classifier is employed for classification. For experimentation, we have extracted 600 word images from each script and total of 3000 word images from video frames. Further, we have made comparative study to show the robustness of symbolic representation and classifier with SVM and ANN classifiers.