| dc.creator |
Myagila, Kasian |
|
| dc.date |
2021-02-01T06:57:45Z |
|
| dc.date |
2021-02-01T06:57:45Z |
|
| dc.date |
2020 |
|
| dc.date.accessioned |
2022-10-20T13:46:59Z |
|
| dc.date.available |
2022-10-20T13:46:59Z |
|
| dc.identifier |
Myagila, K. (2020). A comparative study on performance of support vector machine and convolution neural network on Tanzania sign language translation using image recognition (Master's Dissertation). The University of Dodoma, Dodoma. |
|
| dc.identifier |
http://hdl.handle.net/20.500.12661/2717 |
|
| dc.identifier.uri |
http://hdl.handle.net/20.500.12661/2717 |
|
| dc.description |
Dissertation (MSc Information Technology) |
|
| dc.description |
Sign language has been used by Speech impaired people for communication purposes. Despite being an effective form of communication for speech impaired people, still there is a challenge for people who are unaware of sign language especially those with no such impairment to communicate with speech impaired people. Since Sign Language is a visual based language, several machine learning techniques have been used in sign language translation for better performance results. However, sign languages are different and no study has been found in Tanzania Sign Language, which is a language used by speech impaired people in Tanzania. Moreover, no study has revealed whether there is significant difference in performance between Support Vector Machine and Convolution Neural Network despite the fact that literature show that both have significant performance in different sign languages. This study aimed at comparing the performance of Support Vector Machine and Convolution Neural Network on translating Tanzania Sign Language through image recognition.
The study employed Tanzania Sign Language images as datasets whereby 30 words were chosen from the context of education. The study used dataset of 3000 images that were taken using a camera. To reduce the dimension of datasets, the study adopted Principal Component Analysis to perform feature extraction. Furthermore, the study employed a Combined 5x2cv F test to compare the techniques to determine the significant difference in the performance of the algorithms.
The findings revealed that both techniques have significant rate of both accuracy, precision and recall. Convolution Neural Network scored 96% in all of the parameters while the SVM with Histogram Oriented Gradient feature scored similar rate in precision but lag on recall and accuracy by 1%.
Additionally, the results of using Combined 5x2cv F test yield a p-value of 0.0258 which shows that there is a significant difference in the performance of the two techniques when used to translate the Tanzania Sign Language. Therefore, this study recommends the use of Convolution Neural Network since it has high accuracy and it can provide a significant higher rate of performance compared to Support Vector Machine. |
|
| dc.language |
en |
|
| dc.publisher |
The University of Dodoma |
|
| dc.subject |
Sign language |
|
| dc.subject |
Communication |
|
| dc.subject |
Speech impaired people |
|
| dc.subject |
Visual language |
|
| dc.subject |
Machine learning techniques |
|
| dc.subject |
Convolution Neural Network |
|
| dc.subject |
Vector machine |
|
| dc.subject |
Tanzania |
|
| dc.subject |
Image recognition |
|
| dc.title |
A comparative study on performance of support vector machine and convolution neural network on Tanzania sign language translation using image recognition |
|
| dc.type |
Dissertation |
|