Ship Detection Based on Spatial Partial Features

Wang Hong-hu,Zhou Fu-gen,Jin Ting
DOI: https://doi.org/10.1117/12.2034060
2013-01-01
Abstract:This paper mainly studies how to detect a wide variety of ships from the ship-borne infrared images in order to implement sea monitoring. Different types of ships have significant differences in their appearance. The traditional detection method which uses the global texture features of the object is not suitable to detect varied ships. This paper presents a novel detection algorithm which extracts spatial partial texture features trained by Adaboost to establish the ship model for detection. We first extract all the partial regions of the object through random traversal, and then extract the texture features by using the "Uniform LBP" operator. Compared to the traditional way, we save each partial feature individually as one feature vector, which not only reduces the vector dimension but also highlights the key regions when the partial regions with strong generality are selected by Adaboost at the second step. Finally, the selected partial features are boosted with weights to establish ship model for the ship detection. The proposed approach is efficient and robust in the infrared ship detection.
What problem does this paper attempt to address?