A copula based method for fish species classification

The proposed work develops a method for classification of the species of a fish given in an image, which is a sub-ordinate level classification problem. Sub-ordinate classification is complex as it relies on identifying the notable distinction among the part level characteristics of subjects rather than relying on presence or absence of parts for classification, as done in basic level categorization. Fish image categorization is unique and challenging as the images of same fish species can show significant differences in the fish's attributes when taken in different conditions. Our approach analyses the local patches of images, cropped based on specific body parts, and hence keep comparison more specific to grab more finer details rather than comparing global postures. We have used state-of-the-art multidimensional image descriptor HOG (Histogram of Oriented Gradients) and, colour histograms to create representative feature vectors; feature vectors are summarized using Copula theory which has not been used in many applications in analysing multi-dimensional space despite being one of the most used tools to analyse bivariate data from complex industries like finance and medical science. Our method is very simple yet we have matched the classification accuracy of other proposed complex work for such problems.