基于复杂融合特征与灰度-纹理直方图描述子的红外弱小目标检测追踪算法

Infrared Small Target Detection and Tracking Algorithm Based on Complex Fusion Feature and Gray-texture Histogram Descriptor

为了解决当前红外目标检测追踪算法仅依靠单一图像特征对弱小目标增强,使其在背景杂波与噪声干扰严重条件下,难以剔除图像背景中的伪目标像素,导致弱小目标检测与追踪精度不高,提出了基于复杂融合特征与联合灰度-纹理直方图描述子的红外弱小目标检测与追踪算法。首先,针对红外图像不同特征的背景干扰因素,引入不同方向的腐蚀操作结构元素,设计了分类Top-Hat变换算子,充分抑制背景杂波与噪声,从而将弱小目标从复杂背景中凸显出来;随后,引入方差权重信息熵,构建复杂融合特征,对红外图像进行分割,确定候选目标区域;并基于管道滤波模式,对候选目标区域中的真实弱小目标与伪目标进行筛选,将虚假目标过滤;再考虑弱小目标的强度与纹理特征,基于LBP技术(local binary pattern),设计了灰度-纹理直方图描述子,充分描述红外弱小目标的边缘、线端与角点等鲁棒性特征,较好地保留目标的空域信息,有效剔除图像背景中的伪目标像素;最后,联合均值漂移算法,对红外弱小目标进行精确追踪。实验结果显示:与当前红外目标检测追踪技术相比,在复杂背景干扰条件下,本文算法具有更高的检测精度与更低的追踪误差。

In order to solve the problem of low dim target detection and tracking accuracy induced by enhance the target only by single image feature resulting in difficult to eliminate image background of pseudo target pixels under the serious disturbance condition of clutter wave and noise, the infrared small target detection algorithm based on complex fusion feature and gray-texture histogram descriptor was proposed. Firstly, classified top-hat transform operator was designed by introducing the different corrosion operation structure elements according to the different characteristics of the infrared image background interference factors for adequately suppressing the background clut- ter and noise, and prominentting dim target from the complex background. Then, the region growing technique was introduced to construct the variance weight information entropy, which was used to segment the single IR image and determine the candidate target area. And the real small and weak targets in the candidate target area were identified based on the pipeline filter model, the gray level texture histogram descriptor was designed by fusing the intensity and texture feature of dim target to fully describe the robust feature such as edge and corner points of dim target in infrared image for better retention of spatial information, effectively eliminate the background image in the pseudo target pixel. Finally, the combination of the mean shift algorithm, the infrared dim target was precisely tracked by combining the mean shift algorithm. The experimental results show that this algorithm had higher detection accuracy and lower tracking error in complex background compared with the current infrared target detection and tracking technology.