Abstract—In this paper, we have proposed a novel approach
for handwriting recognition system involving segmentation for
preprocessing steps and using diagonal based feature extraction
technique with neutral network for character recognition. Input
is paragraphs of running text, which is preprocessed to segment
it into normalized individual words. Further, a diagonal based
feature extraction technique is used for extracting the features
of handwritten alphabets. A neural network is trained onto the
dataset containing 55 samples for each of the 26 alphabets for
recognition. A new implicit approach for character recognition
is implemented in this paper which segments a character, into
parts dynamically for character recognition from the text, which
improves the accuracy significantly. A feed forward artificial neural network is being used for character classification, which also
helps in deciding the threshold value for the character separation
from the running text word. The proposed recognition system
performs excellently for separate character written documents
with 100% accuracy. It also performs competitively yielding an
accuracy of 75% for readable non-cursive handwriting and 60%
for cursive handwriting.