New Feature Vector from Freeman Chain Code for Handwritten Roman Character Recognition

This paper proposes features that are extracted solely from Freeman Chain Code (FCC) for handwritten character recognition purpose. Targeting alphanumeric Roman characters, its structure constructed from the chain code is disassembled into segments and landmarks, before each segment is traced to detect predefined line shapes. Two types of feature vectors, sequentially connected shape identifiers and concurrently used shape occurrence counts and size ratios along with landmark positions, are produced from the tracing. Effectiveness of the proposed feature vectors are tested with Hidden Markov Model (HMM) for sequential, while concurrent feature vector is with Artificial Neural Network (ANN), showing mediocre results where only digit character class achieves the highest 80% classification accuracy.