Generic object detection in maritime environment using self-resemblance
Zhihui Zheng, Liping Xiao, Bin Zhou · 2014
We propose a novel unified framework for the initial detection of possible targets within the aerial images using saliency detection. Our method is a bottom-up approach and computes Locally Adaptive Regression Kernel (LARK) from the given image, which measures the likeness of a pixel to its surroundings. Visual saliency is then computed using the self-resemblance measure. The framework results in a saliency map and each pixel indicates the statistical likelihood of saliency of a feature matrix given its surrounding feature matrices. As a similarity measure, matrix cosine similarity is employed. State of the art performance is demonstrated on real aerial images.