Development of text extraction and recognition prototype using Adaptive Resonance Theory 1 (ART1) neural network / Iffarini Idris
Character recognition system can contribute tremendously towards the advancement of automation process and can be useful in many other applications such as Data Entry, Document Processing and Cheque Verification. In this research, a prototype of text extraction and recognition using Adaptive Resonan...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2008
|
| Subjects: | |
| Online Access: | https://ir.uitm.edu.my/id/eprint/65767/1/65767.pdf https://ir.uitm.edu.my/id/eprint/65767/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Character recognition system can contribute tremendously towards the advancement of automation process and can be useful in many other applications such as Data Entry, Document Processing and Cheque Verification. In this research, a prototype of text extraction and recognition using Adaptive Resonance Theory 1 (ARTl) was proposed. For this project, several sets of images were collected from magazines and text books. In prototype design, the interface of ARTl and the ARTl neural network architecture were designed. The pre-process part of this prototype was developed using MATLAB and the recognition part was developed using C++. During the pre-processing stage, images were converted to binary image. Then, the title of the document images was extracted using Mathematical Morphological technique and the characters were segmented using labeling technique. After the pre-processing stage, each of the pixels value that represent the character will be the input to the ARTl network for character recognition process. ARTl neural network has proven to give good performance with 65.7 % recognition rate. A comparative study was conducted between ARTl and Backpropagation Neural Network (BPNN) to compare their recognition performances. BPNN is unable to meet the performance goal because of insufficient number of training data. In conclusion, ARTl is better than BPNN when the number of training data is small. |
|---|