基于小面核滤波的快速红外舰船检测方法

Fast infrared ship detection method based on facet kernel filtering

针对复杂海天背景下红外舰船目标的实时检测需求,提出了一种基于小面核滤波的快速红外舰船检测方法.在频域视觉显著性计算原理的启发下,结合小面拟合模型,设计了一种高效的时域小面卷积核,以抑制背景并增强具有高亮灰度对比边缘的红外舰船目标,进而确定候选目标区域.然后采用一种L0梯度最小化滤波方法,实现候选区域中目标区域的灰度均衡,以完整并准确提取目标形状.最后根据舰船的尺度和形状特性,完成目标判别.实验中对3组公开实拍红外图像数据集进行了测试,实验结果表明:本方法具有较好的鲁棒性,其对3组数据的检测准确率均高于90%,在PC平台上的处理速度维持在10ms量级,优于现有的典型方法,能够满足多数工程应用需求.

Regarding to the requirement of real-time detection for ship targets in infrared images with sea-sky backgrounds,a fast ship detection method based on facet kernel filtering was proposed.Crucially,the spatial facet kernel,which was designed by the facet model according to the theory of visual saliency computation in frequency domain,could effectively extract candidate regions from images by suppressing backgrounds while enhancing ship targets with high intensity relative to surroundings.Then,the L0 gradient minimization filtering was applied to homogenize the targets and backgrounds,respectively,in order to precisely segment the targets.Finally,the judgment on the targets was made according to the ship sizes and shapes.Experimental results on three publicly available infrared image sets demonstrate that the proposed method has good performance on robustness,the detection precision of the proposed method is higher than 90% and the magnitude of the running efficiency is about10 ms on personal computer（PC）platform,which performs better than the state-of-the-art methods and is able to meet the requirements for most practical applications.