A framework for extracting salient local features from 3D models is presented in this paper. In the proposed method, the amount of curvature at a surface point is specified by a positive quantitative measure known as the curvedness. This value is invariant to rigid body transformation such translation and rotation. The curvedness at a surface position is calculated at multiple scales by fitting a manifold to the local neighbourhoods of different sizes. Points corresponding to local maxima and minima of curvedness are selected as suitable features and a confidence measure of each keypoint is also calculated based on the deviation of its curvedness from the neighbouring values. The advantage of this framework is its applicability to both 3D meshes and unstructured point clouds. Experimental results on a different number of models are shown to demonstrate the effectiveness and robustness of our approach.

No abstract

No abstract

No abstract

Sharks are one of the major predators in the ocean. In particular, the great white shark is a primary threat to swimmers. This work proposes an automatic method for the recognition of deformable submerged objects (i.e. sharks) from aerial images of the coast line in an uncontrolled environment. It focuses on great white shark recognition in the surf zone of coastal areas. As the images were taken in an uncontrolled environment and the object shapes of interest are deformable, it is not easy to distinguish sharks from shark-like objects such as dolphins. In this paper, we propose two feature extraction methods that are based on the object's shape: the fish shape feature and shape profile methods. All feature extraction methods are applied to a new image database that contains aerial views of sharks and shark-like objects. The classifiers that are used in our proposed methods are the Support Vector Machine (SVM) and the feed-forward backpropagation neural network.