Autonomous ship recognition from color images

Autonomous ship recognition is an active area for military and commercial applications like harbor surveillance. Accurate identification of unknown contacts is critical in military intelligence. This automated system can help controllers to identify the point of contacts more quickly and accurately. This work mainly focuses on color images attained using digital cameras mounted on ships and harbors. Aside from using digital images for recognition, other information known are distance and course information attained from RADAR. For extracting significant features, spatial pyramid histogram technique is performed on a segmented ship image and support vector machines are used as a classifier. These particular data sets contain 9 different types of ship with 18 different camera angle perspectives for training set, development set and testing set. The training data set contains modeled synthetic images; development and testing data sets contain real images. This work reports two experimental results for ship classification from color images. Our first experiment is based on classification of a synthetic image data set versus real image data set, which means the classifier is trained on the synthetic data set and tested on the real data set, and obtained accuracy is 87.8 percent. Our second experiment is based on classification of synthetic images and real images (combined data set) versus real images, which means the classifier is trained on the combined data-set and tested on a separate real data set, and obtained accuracy is 93.3 percent.