Identifying Genetic Relatedness in Birds Using Visual Patterns

Is there a relationship between visual markings of living organisms and genetic relatedness? Biologists assert that visual markings of animals are preserved through genetic inheritance. However, accurately quantifying genetic relatedness concerning visual marking similarity has not been possible until the advent of machine learning and computer vision algorithms. In this thesis, we set out to answer a question of whether genetic relatedness is correlated to visual marking similarity. We explored several machine learning and computer vision techniques to analyze the problem on bird eggs images having visual markings. We use typical image descriptors such as SIFT, LBP and CNN embeddings to learn the feature vector representation of visual patterns and explore various image similarity algorithms such as siamese neural networks, pairwise SVM and clustering techniques. We identify from the experiment's precision, recall and F-1 scores that none of the models we chose performed well and doesn't show a signal that genetic relatedness can be quantified based on visual markings similarity.