Coral reef image/video classification employing novel octa-angled pattern for triangular sub region and pulse coupled convolutional neural network (PCCNN)

摘要

Coral reef image classification with the help of its texture features is a challenging task, due to its variation in class samples. This is achieved with the proposed feature descriptor termed as Octa-angled Pattern for Triangular sub region (OPT) which selects the neighbor in a triangular pattern in clockwise and counter-clockwise directions. The proposed method reduces the size of feature vector by reducing the bin size of histogram besides improving accuracy. For classification, a novel classifier, named Pulse Coupled Convolutional Neural Network (PCCNN) is employed. The performance of OPT is estimated using F-score. Experiments carried out with a variety of coral images and video data sets, diseased coral data sets and texture data sets to show that OPT technique gets on better than existing feature descriptors. Experimental result shows that the time complexity is reduced and accuracy is improved from 2 to 5% for all coral data sets used.