A robust image enhancement techniques for underwater fish classification in marine environment

Abstract

From literature reviews, the marine environment influences the quality of underwater images and makes the identification of fish species more complex and challenging. The images of the marine environment have low image quality that causes the generated features to be reduced; therefore, this decreases the performance of the classification method. To the best knowledge of the authors, we found out that many researchers have focussed only on determining identification methods without considering the quality of the original data. Therefore, the impact of image enhancement toward the accuracy is yet to be known because this has not been studied comprehensively. To deal with this research gap we propose a new workflow of fish species identification. The workflow for our proposed approach is by using the gray-level co-occurrence matrix (GLCM) feature extraction fed into the back-propagation neural network (BPNN) with contrast-adaptive color correction technique (NCACC) as image enhancements. The experiments demonstrated an improvement in accuracy and kappa measurements for fish species identification from 4.68% to 93.73% and improve from 0.05 to 0.92 respectively. Therefore, our proposed method has the potential to support automatic fish identification systems based on computer vision technology.