Mining of Moving Objects from Time-Series Images and its Application to Satellite Weather Imagery

The framework of mining of moving objects from image data sequence is presented. Scenes are first clustered and labeled by using two-stage SOM that is modified to recognize images including similar moving objects as the same cluster, and that well recognizes scenes including prominent objects. After extraction of images which include prominent objects based on clustering result, the position and the shape of objects are approximated by using mixture gaussian model via EM algorithm, providing the adequate or larger number of components. By adopting the average of the data points in the smaller blocks as the initial parameters, the solutions are stabilized and the identification of components among time-series images and the tracking of a specific object become easier.This framework is applied to a four-year (ranging from 1997 to 2000) dataset of cloud images taken by Japanese weather satellite GMS-5 to evaluate its performance. Modified SOM method well classifies scenes which include prominent moving object, and seasonal variation tendency is detected in the cluster ID sequence. The result of object detection via EM algorithm for summer-type images including clear cloud masses such as typhoons shows that this approach well approximate the adequate distribution of cloud masses in many cases. Objects in the very irregular shapes are also well represented as the mixtures of gaussians.The extracted object information, together with the scene clustering result, is expected to offer us a rich source for knowledge discovery of video datasets. This approach is one of the effective ways of mining video images whose characteristics are unknown in advance, and thus applicable to the various type of applications.     

Salient object detection via local saliency estimation and global homogeneity refinement

This paper presents a new hybrid approach for detecting salient objects in an image. It consists of two processes: local saliency estimation and global-homogeneity refinement. We model the salient object detection problem as a region growing and competition process by propagating the influence of foreground and background seed-patches. First, the initial local saliency of each image patch is measured by fusing local contrasts with spatial priors, thereby the seed-patches of foreground and background are constructed. Later, the global-homogeneous information is utilized to refine the saliency results by evaluating the ratio of the foreground and background likelihoods propagated from the seed-patches. Despite the idea is simple, our method can effectively achieve consistent performance for detecting object saliency. The experimental results demonstrate that our proposed method can accomplish remarkable precision and recall rates with good computational efficiency.     

基于全局注意力的多尺度显著性检测网络

针对现有算法的多尺度特征融合效果不理想和全局信息利用不充分的问题,提出一种基于多尺度优化和全局注意力的显著目标检测模型.利用特征增强模块对从骨干网络VGG-16中提取出来的粗糙特征进行增强,提升特征的显著性表达能力,对不同层次特征融合得到高层级和低层级特征;设计全局注意力模块,利用空洞空间卷积池化金字塔ASPP模块提取...     

A novel approach to combine features for salient object detection using constrained particle swarm optimization

Despite significant amount of research works, the best available visual attention models still lag far behind human performance in predicting salient object. In this paper, we present a novel approach to detect a salient object which involves two phases. In the first phase, three features such as multi-scale contrast, center-surround histogram and color spatial distribution are obtained as described in Liu et al. model. Constrained Particle Swarm Optimization is used in the second phase to determine an optimal weight vector to combine these features to obtain saliency map to distinguish a salient object from the image background. To achieve this, we defined a simple fitness function which highlights a salient object region with well-defined boundary and effectively suppresses the background regions in an image. The performance is evaluated both qualitatively and quantitatively on a publicly available dataset. Experimental results demonstrate that the proposed model outperforms existing state-of-the-art methods in terms of precision, recall, F -measure and area under curve.     

Salient object detection: From pixels to segments

In this paper we propose a novel approach to the task of salient object detection. In contrast to previous salient object detectors that are based on a spotlight attention theory, we follow an object-based attention theory and incorporate the notion of an object directly into our saliency measurements. Particularly, we consider proto-objects as units of the analysis, where a proto-object is a connected image region that can be converted into a plausible object or object-part, once a focus of attention reaches it. As the object-based attention theory suggests, we start with segmenting a complex image into proto-objects and then assess saliency for each proto-object. The most salient proto-object is considered as being a salient object.     

一种基于二维隐马尔可夫模型的图像分类算法

针对图像分块之间的相互依赖关系,提出一种基于二维隐马尔可夫模型的图像分类算 法。该算法将一维隐马尔可夫模型扩展成二维隐马尔可夫模型,模型中相邻的图像分块在平面两个 方向上按条件转移概率进行状态转换,反应出两个维上的依赖关系。隐马尔可夫模型参数通过期望 最大化算法(EM)来估计。同时,本文利用二维Viterbi算法,在训练隐马尔可夫模型的基础上,实现 对图像进行最优分类。文件图像分割的应用表明,隐马尔可夫算法优于CART算法。     

最优算法组合在目标检测图像信号处理中的应用

针对网域内简单背景下目标感知识别处理速度慢、准确率低的问题,提出了基于最优算法组合的目标检测方法;该方法将多种图像处理方法融入目标识别算法中,在采用不同算法对采集到的图像进行预处理、分割和特征提取效果比较的基础上,确定出最优算法组合并设计实验验证其可靠性和鲁棒性;试验结果表明该方法能够快速有效的判别行人目标,识别率达到96.67%,解决了网域内目标检测问题;与其他单一算法相比具有处理速度更快、判别更加有效,实时性高等特点,明显优于一般算法。     

基于语义引导特征聚合的显著性目标检测网络

在显著性目标检测网络的设计中, U型结构使用广泛. 但是在U型结构显著性检测方法中, 普遍存在空间位置细节丢失和边缘难以细化的问题, 针对这些问题, 提出一种基于语义信息引导特征聚合的显著性目标检测网络, 通过高效的特征聚合来获得精细的显著性图. 该网络由混合注意力模块(Mixing attention module, MAM)、增大感受野模块(Enlarged receptive field module, ERFM)和多层次聚合模块(Multi-level aggregation module, MLAM)三个部分组成. 首先, 利用增大感受野模块处理特征提取网络提取出的低层特征, 使其在保留原有边缘细节的同时增大感受野, 以获得更加丰富的空间上/下文信息; 然后, 利用混合注意力模块处理特征提取网络的最后一层特征, 以增强其表征力, 并作为解码过程中的语义指导, 不断指导特征聚合; 最后, 多层次聚合模块对来自不同层次的特征进行有效聚合, 得到最终精细的显著性图. 在6个基准数据集上进行了实验, 结果验证了该方法能够有效地定位显著特征, 并且对边缘细节的细化也很有效.     

改进EM算法在图像检索中的应用

基于相关反馈算法的图像检索经迭代后查询点会陷入局部最优。针对该问题,提出一种基于自适应相关反馈算法的图像检索方法。如果当前查询点达到局部最优,则采用EM裂项算法将该点分解为2个子查询点,通过移动查询点使其各自达到局部最优。如果用户对当前查询不满意,再将这2个子查询点分解为4个子查询点进行处理,以此类推,直到用户满意为止。实验结果表明,与自适应的相关反馈算法、混合反馈算法以及不对称的贝叶斯相关反馈算法相比,该方法的查准率较高。     

视觉Transformer研究的关键问题: 现状及展望

Transformer所具备的长距离建模能力和并行计算能力使其在自然语言处理领域取得了巨大成功并逐步拓展至计算机视觉等领域. 本文以分类任务为切入, 介绍了典型视觉Transformer的基本原理和结构, 并分析了Transformer与卷积神经网络在连接范围、权重动态性和位置表示能力三方面的区别与联系; 同时围绕计算代价、性能提升、训练优化以及结构设计四个方面总结了视觉Transformer研究中的关键问题以及研究进展; 并提出了视觉Transformer的一般性框架; 然后针对检测和分割两个领域, 介绍了视觉Transformer在特征学习、结果产生和真值分配等方面给上层视觉模型设计带来的启发和改变; 并对视觉Transformer未来发展方向进行了展望.     

全局信息引导的多尺度显著物体检测模型

目前方法不能有效提取图像的多尺度特征并完全利用全局信息,对此提出一种新的全局信息引导的多尺度显著物体检测模型.设计了多尺度特征提升模块,提升了各个侧输出特征的多尺度表征能力;利用空洞空间卷积池化金字塔模块提取图像的全局特征;将全局特征直接与各个侧边输出的多尺度特征相融合,利用全局特征引导侧边特征聚焦于目标区域中有用的中...     

基于深度学习的实时图像目标检测系统设计

针对图像目标检测的嵌入式实时应用需求,采用合并计算层的方法对基于MobileNet和单发多框检测器(SSD)的深度学习目标检测算法进行了优化,并采用软硬件结合的设计方法,基于ZYNQ可扩展处理平台设计了实时图像目标检测系统。在系统中,根据优化后的算法设计了一款多处理器核的深度学习算法加速器,并采用PYTHON语言设计了系统的软件。经过多个实验测试,深度学习目标检测系统处理速度可以达到45FPS,是深度学习软件框架在CPU上运行速度的4.9倍,在GPU上的1.7倍,完全满足实时图像目标检测的需求。     

利用SVM改进Adaboost算法的人脸检测精度

提出利用SVM分类方法改进Adaboost算法的人脸检测精度。该方法先通过Adaboost算法找出图像中的候选人脸区域,根据训练样本集中的人脸和非人脸样本训练出分类器支持向量机(SVM),然后通过SVM分类器从候选人脸区域中最终确定人脸区域。实验结果证明,SVM分类算法可以提高检测精度,使检测算法具有更好的检测效果。     

一种新的基于目标和特征的图像分类框架

首先提出了一种图像分类的结构。根据子块技术和形状描述技术来获取图像中的目标,如果图像类属于目标图像类,则提取图像中目标图像的特征,否则提取整幅图像的底层特征。然后利用主分量分析方法（PCA）对提取的特征进行降维处理,降维处理后的数据由支持向量机进行分类。该方法在标准的Corel图像库上进行了测试,实验结果表明提出的方法有效地提高了图像分类的性能,图像分类的结果与图像的高层语义概念相一致。     

支持向量机分类器遥感图像分类研究

SVM分类器核函数的选择以及参数的设置直接影响系统的泛化能力和运行速度。引入交叉验证技术和栅格搜索技术,对径向基核、多项式核和Sigmoid核函数应用于图像多类别分类的性能进行理论推导、测试及分析,求得三种核函数应用于SVM分类器的性能,并证明了栅格搜索寻找最优参数的有效性。最后通过对TM 6波段BSQ格式遥感图像进行分类对比证明了SVM分类器核函数用于TM图像分类的可行性及高效性。     

卷积神经网络在图像分类和目标检测应用综述

卷积神经网络具有强大的特征学习能力,随着大数据时代的到来和计算机能力的提升,近年来卷积神经网络在图像识别、目标检测等领域取得了突破性进展,掀起了新的研究热潮。综述卷积神经网络的基本原理,以及其在图像分类、目标检测上的研究进展和典型模型,最后分析了卷积神经网络目前的问题,并展望了未来的发展方向。     

基于颜色直方图的EM对象跟踪算法

基于均值漂移运动目标跟踪的迭代算法,简单可靠,可以方便准确的找到一个基于内核的概率密度函数估计目标的位置。但是该算法对目标尺寸形状变化的适应能力比较差,文章提出一个改进的均值漂移算法。新算法同时估计目标的位置和用协方差矩阵来描述的目标的形状,能够处理对象的角度和形状大小发生变化时的跟踪问题,运用新的算法实现以颜色直方图为基础的非刚性目标跟踪算法。实验表明,该改进的算法在不同环境下跟踪目标的鲁棒性很好,尤其对跟踪目标的形状和尺寸的改变,具有很强的适用性。     

基于熵的运动目标检测

目前，运动目标的检测是计算机视觉领域中最活跃的研究主题之一。本文介绍了光流法、帧间差分等运动目标的检测方法，提出了一种基于熵的运动目标检测方法，实验结果表明了该方法的鲁棒性和有效性。