Isolated handwritten digit recognition using genetic algorithm / Farizatul Sarina Mohd Sari
There are various types of recognition available that are face recognition, digit recognition and many more. Various techniques used to recognize pattern such as Neural Network (NN), Genetic Algorithm (GA) and others. This project focuses on isolated digit recognition using GA as the recognition tec...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2007
|
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | There are various types of recognition available that are face recognition, digit recognition and many more. Various techniques used to recognize pattern such as Neural Network (NN), Genetic Algorithm (GA) and others. This project focuses on isolated digit recognition using GA as the recognition technique. The objective of the project is to determine the best performance rate between two types of images that are 16 x 16 images and 32 x 32 images with different type of encoding scheme and crossover rate and also to determine the best crossover rates between 0.6,0.7 and O.S. The project is based on a series of experiments that compare performance rate to recognize digit using GA. There are three sources of data involved in the project that are using MNIST database, USPS database and researcher's own data. The data will be in a form of bitmap pictures with 16 x 16 images and 32 x 32 images. In the algorithm, two types of encoding scheme used that are binary encoding and permutation encoding. The type of crossover used is one-point crossover with multiple rates that are 0.6, 0.7 and 0.8 and the type mutation used is inversion with the rate of 0.01. There are four types of experiment that have been conducted. Firstly, 16 x 16 images represented in binary encoding with different crossover rates. Secondly, 16 x 16 images represented in permutation encoding with different crossover rates. Thirdly, 32 x 32 images represented in binary encoding with different crossover rates. Lastly, 32 x 32 images represented in permutation encodmg with different crossover rates. From the experiment made, it was showed that the best performance to recognize a digit is S hours and 41 minutes for 16 x 16 images using
binary encoding and 0.7 crossover rate. The best crossover rate among the three rates that is being tested is 0.7. The result of this research shows that the performance rate can be improved so that it can be implemented in various field. |
|---|