K-L transform-based method for detecting saliency of small object on sea

Abstract

The invention discloses a K-L transform-based method for detecting saliency of a small object on the sea. The method comprises: S1, extracting a brightness channel feature graph and two color channel feature graphs in seaborne scene visible image CIELab space; S2, carrying out Fourier transform on all channels obtained by the S1 to obtain amplitude spectra and phase spectra; S3, carrying out K-L conversion on the amplitude spectra, obtained at the S2, of all channels; S4, selecting main components of K-L transform results, obtained at the S3, of the amplitude spectra of the respective channels according to features of a seaborne image data set, and calculating a differential spectrum matrix of all amplitude spectra; S5, combining the differential spectrum matrix, obtained at S4, of all channels with all original amplitude spectra to carry out inverse discrete Fourier transform, thereby obtaining saliency maps of all channels; and S6, calculating a total saliency map by using a dynamic weighting method. According to the invention, small targets are highlighted and a background is suppressed, so that small target detection and tracking on the sea are realized.

Description

technical field [0001] The invention relates to the fields of computer vision and image processing, in particular to a method for detecting the saliency of small sea targets based on Karhunen-Loève (K-L) transformation. Background technique [0002] The detection and tracking of small targets in maritime scenes has always been a difficult point in the field of computer vision, and the detection and tracking of such targets is a key issue in the fields of maritime video surveillance and shipwreck search and rescue. [0003] In the past two decades, the research on the human visual attention mechanism has been a hot topic in the field of computer vision and image processing, and has been applied in many technical fields. Saliency detection is one of the core issues of its research, which mainly revolves around how to measure and calculate saliency from two perspectives of image and video features or combined with prior information. Saliency detection methods based on image fe...

AI Technical Summary

Problems solved by technology

[0004] However, the main problems of the existing methods are: the performance of the algorithm is constantly improving, but the algorithm is complex and not easy to apply in real time; it mainly solves the problem of saliency detection of large targets in natural scenes, and only some scholars have discussed the classification of target data of different sizes. There are relatively few studies on saliency detection algorithms for maritime scenes; the prior information applied in some algorithms is not suitable for small targets in maritime scenes, for example, the central bias constraint of natural scenes is often not applicable in maritime scenes

Images