基于背景先验与中心先验的显著性目标检测北大核心CSCD

针对传统的图流行排序显著性目标检测算法存在先验信息单一,显著目标检测不完整的问题,提出一种新的基于背景先验与中心先验的显著性目标检测算法。首先将图像边界节点作为背景种子进行流行排序获得粗略的前景区域,将其再次流行排序得到初步显著图;然后利用Harris角点检测、聚类实现中心先验显著性检测,捕获中心显著信息;最后在初步显著图上融合图像中心显著性,得到最终显著图。本文对综合指标、精确率-召回率曲线、F-measure值以及平均绝对误差(mean absolute error,MAE)值进行实验评估,在公开数据集MSRA-10K和ECSSD上进行的实验结果表明:对比10种主流算法,本文算法在不同的评估指标上都具有较好的表现,且能准确地突出显著目标,提升背景抑制效果。     

Saliency guided self-attention network for pedestrian attribute recognition in surveillance scenarios

Pedestrian attribute recognition is often considered as a multi-label image classification task. In order to make full use of attribute-related location information, a saliency guided sel-attention network ( SGSA-Net) was proposed to weakly supervise attribute localization, without annotations of attribute-related regions. Saliency priors were integrated into the spatial attention module ( SAM ). Meanwhile,channel-wise attention and spatial attention were introduced into the network. Moreover, a weighted binary cross-entropy loss ( WCEL) function was employed to handle the imbalance of training data. Extensive experiments on richly annotated pedestrian ( RAP) and pedestrian attribute ( PETA) datasets demonstrated that SGSA-Net outperformed other state-of-the-art methods.     

孪生网络框架下融合显著性和干扰在线学习的航拍目标跟踪算法

针对一般跟踪算法不能很好地解决航拍视频下目标分辨率低、视场大、视角变化多等特殊难点,该文提出一种融合目标显著性和在线学习干扰因子的无人机(UAV)跟踪算法.通用模型预训练的深层特征无法有效地识别航拍目标,该文跟踪算法能根据反向传播梯度识别每个卷积滤波器的重要性来更好地选择目标显著性特征,以此凸显航拍目标特性.另外充分利用连续视频丰富的上下文信息,通过引导目标外观模型与当前帧尽可能相似地来在线学习动态目标的干扰因子,从而实现可靠的自适应匹配跟踪.实验证明:该算法在跟踪难点更多的UAV123数据集上跟踪成功率和准确率分别比孪生网络基准算法高5.3%和3.6%,同时速度达到平均28.7帧/s,基本满足航拍目标跟踪准确性和实时性需求.     

基于多流网络一致性的视频显著性检测

现有的视频显著性检测算法通常采用双流结构提取视频的时空线索,其中运动信息作为双流结构的一个分支,在显著物体发生剧烈或慢速移动时存在运动估计准确率低的问题,并且不合理的训练数据或方案使得权重偏向单个分支结构。提出一种基于多流网络一致性的视频显著性检测算法MSNC。设计并使用一种新的三重网络结构提取预选目标区域的颜色信息、时序信息和先验特征,通过先验特征补偿运动流的缺陷,并提高运动线索的利用率。采用多流一致性融合模型优化三流分支,得到不同特征的最佳融合方案。同时通过循环训练策略平衡三重网络的权重,以避免网络过度拟合单流分支,从而有效地提高运动估计和定位的准确率。在Davis数据集上的实验结果表明,相比PCSA、SSAV、MGA等算法,该算法的鲁棒性更优,其maxF和S-Measure值分别达到0.893和0.912,MAE仅为0.021。     

结合l1/2范数与显著性约束的背景减除

传统背景减除模型在背景静止和前景对象移动较快时提取到的前景效果较好,但当背景变化或前景对象移动缓慢时容易将动态背景误判为前景或检测出的前景有较多空洞。针对传统背景减除模型在动态背景和前景对象移动缓慢条件下存在前景检测精度下降的问题,提出一种基于l_1/2范数与显著性约束的背景减除模型。将观测数据分为低秩背景、运动前景和动态干扰3类,利用l_1/2范数约束运动前景加强前景稀疏性,有效抑制动态背景对前景提取造成的干扰,提高运动前景在动态背景中的检测精度。引入视频每一帧的显著性约束,通过对每一帧图像进行低秩稀疏分解来检测移动缓慢的目标。实验结果表明,该模型对于复杂场景具有较强的适应能力,可有效去除动态背景对前景的干扰,快速检测出移动缓慢的前景对象,相比于l_1/1/2-RPCA背景减除模型的平均查全率、查准率和调和平均值分别提升了9、14和10个百分点。     

Efficient random saliency map detection

Most image retargeting algorithms rely heavily on valid saliency map detection to proceed. However, the inefficiency of high quality saliency map detection severely restricts the application of these image retargeting methods. In this paper, we propose a random algorithm for efficient context-aware saliency map detection. Our method is a multiple level saliency map detection algorithm that integrates multiple level coarse saliency maps into the resulting saliency map and selectively updates unreliable regio...     

Using a Visual Attention Model to Improve Gaze Tracking Systems in Interactive 3D Applications

This paper introduces the use of a visual attention model to improve the accuracy of gaze tracking systems. Visual attention models simulate the selective attention part of the human visual system. For instance, in a bottom‐up approach, a saliency map is defined for the image and gives an attention weight to every pixel of the image as a function of its colour, edge or intensity. Our algorithm uses an uncertainty window, defined by the gaze tracker accuracy, and located around the gaze point given by the tracker. Then, using a visual attention model, it searches for the most salient points, or objects, located inside this uncertainty window, and determines a novel, and hopefully, better gaze point. This combination of a gaze tracker together with a visual attention model is considered as the main contribution of the paper. We demonstrate the promising results of our method by presenting two experiments conducted in two different contexts: (1) a free exploration of a visually rich 3D virtual environment without a specific task, and (2) a video game based on gaze tracking involving a selection task. Our approach can be used to improve real‐time gaze tracking systems in many interactive 3D applications such as video games or virtual reality applications. The use of a visual attention model can be adapted to any gaze tracker and the visual attention model can also be adapted to the application in which it is used.     

面向视频压缩的显著性协同检测JND模型

为了更好的将人眼感知特性用于视频压缩系统,提出了一种改进的基于显著性协同检测的恰可察觉失真模型（Just Noticeable Distortion,JND）.该模型通过像素域和变换域下联合建模计算得到的最优JND模型,基于上下文感知的显著性算法得到相应的显著图,并将检测结果用于JND模型权值分配.提出的JND残差滤波器可以嵌入到HEVC视频编码框架中.实验结果表明:在全I帧配置下,提出的算法编码结果与HM16相比,在视觉主观感知质量一致的情况下,平均码率可节省10.7%.     

面向多种场景的视频对象自动分割算法

针对当前应用于视频对象分割的图割方法容易在复杂环境、镜头移动、光照不稳定等场景下鲁棒性不佳的问题,提出了结合光流和图割的视频对象分割算法.主要思路是通过分析前景对象的运动信息,得到单帧图像上前景区域的先验知识,从而改善分割结果.论文首先通过光流场采集视频中动作信息,并提取出前景对象先验区域,然后结合前景和背景先验区域建立图割模型,实现前景对象分割.最后为提高算法在不同场景下的鲁棒性,本文改进了传统的测地显著性模型,并基于视频本征的时域平滑性,提出了基于混合高斯模型的动态位置模型优化机制.在两个标准数据集上的实验结果表明,所提算法与当前其他视频对象分割算法相比,降低了分割结果的错误率,有效提高了在多种场景下的鲁棒性.     

基于说话人辨识的自上而下听觉显著性注意模型

为体现听觉注意神经信息处理计算机制对听觉场景内容的自动分析与理解功能,本文基于人耳对频率变换的感知特性,结合深度信念网络的说话人辨识与听觉显著模型,提出了一种自上而下的听觉显著性注意提取模型.仿真结果表明:该模型具有可行性,同时在利用深度信念网络的说话人辨识技术中能够有效地凸显目标说话人的显著度.