Performance Evaluation Of Multiscale Covariance Descriptor In Underwater Object Detection

Object detection is the fundamental process for the majority of the investigation projects in the submarine environment, and object detection is mainly based on image description done by the appropriate descriptor. In this paper we select and optimize parameters of multi-scale covariance descriptor for object detection in the submarine context. We adapt the descriptor parameters to be suitable to cope with the degradation of image quality in underwater environment, working on the homogeneity error tolerance and the precision degree of description. We justify the use of specific parameters values and well defined features. To perform our work we use support vector machine for data classification and Maris dataset as a benchmark.