摘要
针对复杂背景的图像中纹理目标检测实时性低,准确率不高的问题,提出了一种基于梯度响应图和视觉显著性的快速实时目标检测方法。首先,利用"BING"特征算法快速搜索图像中目标,确定目标大概位置,同时去除冗余背景;然后,引入视觉注意机制,利用改进谱残差模型消除目标纹理响应,同时加强目标轮廓的梯度显著性,构建梯度显著图;最后,在梯度显著图的基础上,建立梯度响应图,利用模板主方向和查找表的实时模板匹配法,通过计算梯度响应图与模板之间梯度主方向的相似性度量值进行目标检测。在"ECSSD"公开数据集上进行测试,该算法能够在复杂背景下,快速实时地完成目标检测,且针对纹理目标具有很高的鲁棒性。实验结果表明,该方法检测准确率高达95.3%,具有良好的检测性能。
A fast real-time target detection method based on gradient response graph and visual saliency is proposed.Firstly,the BING feature algorithm is used to quickly search the target in the image,determine the approximate location of the target,and remove the redundant background.Then,this paper introduces the visual attention mechanism,use the improved spectral residual model to eliminate the target texture response,and enhance the gradient saliency of the target contour to build a gradient saliency map.Finally,based on the gradient saliency map,a gradient response map is established.Using the real-time template matching method of the template principal direction and the look-up table,the target detection is performed by calculating the similarity measure of the gradient principal direction between the gradient response map and the template.
出处
《工业控制计算机》
2019年第5期85-87,共3页
Industrial Control Computer
基金
国家自然科学基金(61171058)
关键词
目标检测
梯度响应图
显著性
模板匹配
相似性度量
target detection
gradient response graph
significance
template matching
similarity measure