Unsupervised image segmentation for fish egg images
Fisheries sector is still actively developing in most countries. Fish is one of most important source of human diet since it contain high amount of protein and fat. Fish aquaculture requires detailed and delicate process. The egg counting process is purposely to check and maintain the quality of the...
Saved in:
| Main Author: | |
|---|---|
| Format: | Student Project |
| Language: | en |
| Published: |
2017
|
| Subjects: | |
| Online Access: | https://ir.uitm.edu.my/id/eprint/132939/1/132939.pdf https://ir.uitm.edu.my/id/eprint/132939/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Fisheries sector is still actively developing in most countries. Fish is one of most important source of human diet since it contain high amount of protein and fat. Fish aquaculture requires detailed and delicate process. The egg counting process is purposely to check and maintain the quality of the females. This process occurred to distinguish the good and poor quality of female. At present, the counting process is conducted manually by operator. The staff captures images of fish egg. Then, the counting process is performed manually based on the fish image displayed on the computer monitor. This manual counting process is time consuming, eye strain and usually prone to error. The use of image processing for fish egg counting is relatively new technique that can help to improve the accuracy and efficiency of counting process. Image segmentation is a main part in image processing based-fish egg counting. The success of the final counting process depends mainly on the performance of image segmentation. Therefore this thesis investigates the performance of three unsupervised image segmentation methods for tilapia egg image segmentation. The unsupervised methods are k-mean clustering, watershed and Otsu thresholding. First, a total of twenty tilapia egg images with size of 1280 x 1280 is acquired from a private tilapia hatchery, Aquadesa Resources Sdn. Bhd. and used in the experiment. The tilapia egg images are segmented to remove the unwanted background. For the segmentation technique, the proposed method is based on gray level image for computational simplicity. The gray level intensities are fed into the k- mean clustering, watershed and Otsu’s thresholding to perform the segmentation process. The results of the proposed method are evaluated using Jaccard index to evaluate its effectiveness in segmenting tilapia egg images. Simulation results indicated that the watershed algorithm outperformed the other two methods with average of 99.62% and 99.65% for clear and blur tilapia egg images, respectively. |
|---|