Infrared Sea-Sky Line Detection Utilizing Self-Adaptive Laplacian of Gaussian Filter and Visual-Saliency-Based Probabilistic Hough Transform

Abstract

Infrared (IR) sea-sky line (SSL) detection is an important assisting method for maritime monitoring. In complex sea-sky backgrounds, strong clouds and waves often have IR features similar to the real SSL, which challenges the true positive rate (TPR) of SSL detection. We propose a new SSL detection algorithm based on a novel self-adaptive Laplacian of Gaussian (SALoG) filter and visual-saliency-based probabilistic Hough transform (VSBPHT). The Laplacian filter of the SALoG has a self-adaptive new kernel, which is calculated by the overall gray information of the image. Its proportional-gray-reduction feature can weaken the interferences, especially in the sky region, and relatively enhance the SSL simultaneously, so that we can detect the SSL from the top to the bottom of the image which could avoid the disturbances in the complex sea region. Also, in order to improve the processing speed and accuracy, the algorithm uses probabilistic Hough transform (PHT) to extract the candidate line segment set and refine it by SSL visual saliency features, the final segments are then used to fit the SSL. Experimental results based on three IR image sequences with different backgrounds show that the proposed algorithm has the highest TPR and precision compared with state-of-the-art algorithms, and the average time consumption could meet the needs for real-time processing as well.