Offline cursive script word recognition : a survey
read more
Citations
An overview of character recognition focused on off-line handwriting
Offline recognition of unconstrained handwritten texts using HMMs and statistical language models
Improving Offline Handwritten Text Recognition with Hybrid HMM/ANN Models
A survey on off-line Cursive Word recognition
Markov models for offline handwriting recognition: a survey
References
A tutorial on hidden Markov models and selected applications in speech recognition
Dynamic Programming
Binary codes capable of correcting deletions, insertions, and reversals
Fundamentals of speech recognition
Related Papers (5)
Online and off-line handwriting recognition: a comprehensive survey
Frequently Asked Questions (13)
Q2. What is the basic concept of a complete word recognition process?
A typical complete word recognition process consists of the following parts: preprocessing, a possible segmentation or fragmentation, feature extraction, recognition, and post-processing.
Q3. How can the dynamic-programming phase be restored?
After the dynamic-programming phase is finished, the optimal segmentation that is associated with the highest score can be restored by back tracing.
Q4. What is the term primitive with respect to segments?
The term primitive with respect to segments stands for elementary segments that were created by the segmentation algorithm that was used.
Q5. Why are there no backwards transitions in a sub-HMM?
Because of the nature of handwriting that dictates a left to right movement, there are no backwards transitions in a sub-HMM representing a letter.
Q6. Why are HMMs more easily trained and adapted than dynamic programming methods?
Due to the well-known Baum-Welch algorithm, the HMMs are more easily trained and adapted than simple dynamic-programming methods based on minimum edit-distance.
Q7. Why did HMMs gain popularity in classification problems?
HMMs gained their popularity in classification problems in general and cursive word recognition in particular because they are most suitable to model the variance that usually appears in symbolic description chains of objects.
Q8. How is the symbolic description chain produced?
The symbolic description chain is produced, like in some segmentation-based methods, by translating each feature vector that might be found in a segment into a unique symbol.
Q9. What is the process used to validate or reject active words?
The top-down process is used to validate or reject active words, according to the existence or absence, respectively, of features that are required to create the missing letters.
Q10. What is the minimum edit distance between two chains of symbols?
The minimum edit-distance between two chains of symbols denoted by o1 . . . om and r1 . . . rn is the cheapest cost one needs to pay in order to transform one chain to the other using the operations of symbol deletion, substitution, or insertion of an extra one.
Q11. What is the way to find the match between blocks of primitive segments?
In segmentation-based methods the recognition process is based on an attempt to find the best complete bipartite match between blocks of primitive segments and a word’s letters.
Q12. What is the main criterion for classifying the recognition method used?
In this paper the authors have decided to use the type of the segmentation scheme as the major criterion for classifying the recognition method used.
Q13. What is the bipartite matching method?
When comparing the observations derived from a word image with a similar reference of a lexicon word, one attempts to find the best bipartite matching between the two, using a dynamic-programming algorithm.