Development of methodology for Malaysian sign language recognition
Development of Methodology for Malaysian Sign Language Recognition is undertaken on Malaysian Sign Language (MSL) aimed to reduce the challenge faced by the deaf community, while communicating in a normal society by providing a mechanism to translate sign language into words. Therefore, in this thes...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2013
|
| Subjects: | |
| Online Access: | https://eprints.ums.edu.my/id/eprint/41796/1/24%20PAGES.pdf https://eprints.ums.edu.my/id/eprint/41796/2/FULLTEXT.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Development of Methodology for Malaysian Sign Language Recognition is undertaken on Malaysian Sign Language (MSL) aimed to reduce the challenge faced by the deaf community, while communicating in a normal society by providing a mechanism to translate sign language into words. Therefore, in this thesis, the main focus is on vision-based hand gesture recognition through investigation of each step including frame selection, skin segmentation, trajectory tracking and recognition. There were 112 isolated MSL chosen; each of the sign is performed six times by different signers, making six video sequence databases of MSL. From the six databases, five databases are used for training and one database is used for testing. In frame selection, only selected frames from the video are required to represent the respected signs. Then, skin likelihood model is built first to filter out skin region, which consists of hands and face region from non-skin region. Centroids, distances and orientations of hand skin region are collected as feature vectors. When overlapping between hand-face or hand-hand occurred, Linear Kalman Filter is used to distinguish the separate features. Each of the feature vectors is translated into chain code for the recognition stage using Hidden Markov Model. There will be six maximum logarithmic probability values that are produced representing the respective signs. In the testing stage, these six values need to be summed up in order to check the correct classification value, which is based on the highest logarithmic probability values. The recognition rate using HMM for testing data achieved up to 83.1%. |
|---|