显著中心先验和显著-深度概率矫正的RGB-D显著目标检测

随着深度特征在图像显著检测领域中发挥越来越重要的作用,传统的RGB图像显著检测模型由于未能充分利用深度信息已经不能适用于RGB-D图像的显著检测。该文提出显著中心先验和显著-深度(S-D)概率矫正的RGB-D显著检测模型,使得深度特征和RGB特征间相互指导,相互补充。首先,依据3维空间权重和深度先验获取深度图像初步显著图;其次,采用特征融合的流形排序算法获取RGB图像的初步显著图。接着,计算基于深度的显著中心先验,并以该先验作为显著权重进一步提升RGB图像的显著检测结果,获取RGB图像最终显著图;再次,计算显著-深度矫正概率,并对深度图的初步显著检测结果使用此概率进行矫正。接着,计算基于RGB的显著中心先验,并以该先验作为显著权重进一步提升深度图像矫正后的显著检测结果,获取深度图像的最终显著图;最后,采用优化框架对深度图像最终显著图进行优化得到RGB-D图像最终的显著图。所有的对比实验都是在公开的数据集NLPR RGBD-1000数据集上进行,实验结果显示该文算法较当前流行的算法有更好的性能。     

融合显著深度特征的RGB-D图像显著目标检测

深度信息被证明是人类视觉的重要组成部分,然而大部分显著性检测工作侧重于2维图像上的方法,并不能很好地利用深度进行RGB-D图像显著性检测。该文提出一种融合显著深度特征的RGB-D图像显著目标检测方法,提取基于颜色和深度显著图的综合特征,根据构图先验和背景先验的方法进行显著目标检测。首先,对原始深度图进行预处理：使用背景顶点区域、构图交点和紧密度处理深度图,多角度融合形成深度显著图,并作为显著深度特征,结合颜色特征形成综合特征;其次,从前景角度,将综合特征通过边连接权重构造关联矩阵,根据构图先验,假设多层中心矩形为前景种子,通过流形排序方法计算出RGB-D图像的前景显著图;从背景角度,根据背景先验以及边界连通性计算出背景显著图;最后,将前景显著图和背景显著图进行融合并优化得到最终显著图。实验采用RGB-D1000数据集进行显著性检测,并与4种不同的方法进行对比,所提方法的显著性检测结果更接近人工标定结果,PR(查准率-查全率)曲线显示在相同召回率下准确率高于其他方法。     

基于多模态特征融合监督的RGB-D图像显著性检测

RGB-D图像显著性检测是在一组成对的RGB和Depth图中识别出视觉上最显著突出的目标区域。已有的双流网络,同等对待多模态的RGB和Depth图像数据,在提取特征方面几乎一致。然而,低层的Depth特征存在较大噪声,不能很好地表征图像特征。因此,该文提出一种多模态特征融合监督的RGB-D图像显著性检测网络,通过两个独立流分别学习RGB和Depth数据,使用双流侧边监督模块分别获取网络各层基于RGB和Depth特征的显著图,然后采用多模态特征融合模块来融合后3层RGB和Depth高维信息生成高层显著预测结果。网络从第1层至第5层逐步生成RGB和Depth各模态特征,然后从第5层到第3层,利用高层指导低层的方式产生多模态融合特征,接着从第2层到第1层,利用第3层产生的融合特征去逐步地优化前两层的RGB特征,最终输出既包含RGB低层信息又融合RGB-D高层多模态信息的显著图。在3个公开数据集上的实验表明,该文所提网络因为使用了双流侧边监督模块和多模态特征融合模块,其性能优于目前主流的RGB-D显著性检测模型,具有较强的鲁棒性。     

基于深度特征提取和图神经网络匹配的图像复制粘贴篡改检测

针对图像中特征提取不均匀、单尺度超像素划分对伪造定位结果影响较大的问题,提出一种基于深度特征提取和图神经网络(graph neural network,GNN) 匹配的图像复制粘贴篡改检测(cope-move forgery detection,CMFD) 算法。首先将图像进行多尺度超像素分割并提取深度特征,为保证特征点数目充足,以超像素为单位计算特征点分布的均匀度,自适应降低特征提取阈值;随后引入新的基于注意力机制的GNN特征匹配器,进行超像素间的迭代匹配,且用随机采样一致性(random sample consensus,RANSAC) 算法消除误匹配;最后将多尺度匹配结果进行融合,精确定位篡改区域。实验表明,所提算法具有良好的性能,也证明了GNN在图像篡改检测领域的可用性。     

基于提取双选紧密特征的RGB-D显著性检测网络

针对现有算法对不同来源特征之间的交互选择关注度欠缺以及对跨模态特征提取不充分的问题,提出了一种基于提取双选紧密特征的RGB-D显著性检测网络。首先,为了筛选出能够同时增强RGB图像显著区域和深度图像显著区域的特征,引入双向选择模块(bi-directional selection module, BSM);为了解决跨模态特征提取不充分,导致算法计算冗余且精度低的问题,引入紧密提取模块(dense extraction module, DEM);最后,通过特征聚合模块(feature aggregation module, FAM)对密集特征进行级联融合,并将循环残差优化模块(recurrent residual refinement aggregation module, RAM)配合深度监督实现粗显著图的持续优化,最终得到精确的显著图。在4个广泛使用的数据集上进行的综合实验表明,本文提出的算法在4个关键指标方面优于7种现有方法。     

基于混合卷积自编码极限学习机的RGB-D物体识别

有效学习丰富的表征信息在RGB-D目标识别任务中至关重要,是实现高泛化性能的关键。针对卷积神经网络训练时间长的问题,提出了一种混合卷积自编码极限学习机（HCAE-ELM）结构,包括卷积神经网络（CNN）和自编码极限学习机（AE-ELM）,该结构合并了CNN的有效性和AE-ELM快速性的优点。它使用卷积层和池化层分别从RGB和深度图来有效提取低阶特征,然后在共享层合并两种模型特征,输入到自编码极限学习机中以得到高层次的特征,最终的特征使用极限学习机（ELM）进行分类,以获得更好的快速泛化能力。文中在标准的RGB-D数据集上进行了评估测试,其实验结果表明,相比较深度学习和其他的ELM方法,文中的混合卷积自编码极限学习机模型取得了良好的测试准确率,并且有效地缩减了训练时间。     

基于跨模态特征融合的RGB-D显著性目标检测

显著性目标检测(SOD)作为目前计算机视觉以及计算机图形学领域中研究的基本课题之一,是许多其他复杂任务的预处理阶段的任务,对例如图像理解与解释、视觉追踪、语义分割,视频分析等对象级应用的发展起到了极大的推动作用。随着深度传感器的普及,深度图像中蕴含的空间信息线索在显著性检测研究中提供了与RGB图像中蕴含的不同模态的辅助补充特征信息,这对于检测精度的提升来说愈发重要,因此如何有效地融合RGB与深度图像中的不同模态间的特征信息成为了RGB-D显著性目标检测课题中研究的重要问题。针对RGB与Depth模态间的特征融合问题,本文设计了一种基于跨模态特征信息融合的双流RGB-D显著目标检测网络模型,通过使用设计的跨模态特征融合模块去除某些低质量深度图带入的冗余与噪音,随后提取放大被优化改良过后的深度特征线索与RGB特征线索间的相似性与差异性,完成跨模态特征信息的有效融合。除此之外在网络编码结构的顶端增加了改良的非局部模块,通过自注意力机制更好地捕捉了的上下文信息以及像素间的长距离依赖。通过使用的两个数据集上的实验表明,这一模型在4个评价指标上取得了较好的表现。     

基于特征转移金字塔网络的SAR图像跨尺度目标检测

SAR图像多尺度目标检测能够实现大场景SAR图像中关键目标的定位与识别,是SAR图像解译的关键技术之一.然而针对尺寸相差较大的SAR目标的同时检测,即跨尺度目标检测问题,现有目标检测方法难以实现.该文提出一种基于特征转移金字塔网络(FTPN)的SAR图像跨尺度目标检测方法.在特征提取阶段采用特征转移方法,实现各层特征图...     

基于卷积神经网络与全局优化的协同显著性检测

针对目前协同显著性检测问题中存在的协同性较差、误匹配和复杂场景下检测效果不佳等问题,该文提出一种基于卷积神经网络与全局优化的协同显著性检测算法。首先基于VGG16Net构建了全卷积结构的显著性检测网络,该网络能够模拟人类视觉注意机制,从高级语义层次提取一幅图像中的显著性区域;然后在传统单幅图像显著性优化模型的基础上构造了全局协同显著性优化模型。该模型通过超像素匹配机制,实现当前超像素块显著值在图像内与图像间的传播与共享,使得优化后的显著图相对于初始显著图具有更好的协同性与一致性。最后,该文创新性地引入图像间显著性传播约束因子来克服超像素误匹配带来的影响。在公开测试数据集上的实验结果表明,所提算法在检测精度和检测效率上优于目前的主流算法,并具有较强的鲁棒性。     

基于超像素级卷积神经网络的多聚焦图像融合算法

该文提出了基于超像素级卷积神经网络(sp-CNN)的多聚焦图像融合算法。该方法首先对源图像进行多尺度超像素分割,将获取的超像素输入sp-CNN,并对输出的初始分类映射图进行连通域操作得到初始决策图;然后根据多幅初始决策图的异同获得不确定区域,并利用空间频率对其再分类,得到阶段决策图;最后利用形态学对阶段决策图进行后处理,并根据所得的最终决策图融合图像。该文算法直接利用超像素分割块进行图像融合,其相较以往利用重叠块的融合算法可达到降低时间复杂度的目的,同时可获得较好的融合效果。     

Attention-based contextual interaction asymmetric network for RGB-D saliency prediction

Saliency prediction on RGB-D images is an underexplored and challenging task in computer vision. We propose a channel-wise attention and contextual interaction asymmetric network for RGB-D saliency prediction. In the proposed network, a common feature extractor provides cross-modal complementarity between the RGB image and corresponding depth map. In addition, we introduce a four-stream feature-interaction module that fully leverages multiscale and cross-modal features for extracting contextual information. Moreover, we propose a channel-wise attention module to highlight the feature representation of salient regions. Finally, we refine coarse maps through a corresponding refinement block. Experimental results show that the proposed network achieves a performance comparable with state-of-the-art saliency prediction methods on two representative datasets.     

基于鲁棒前景选择的显著性检测

显著性检测是指自动提取未知场景中符合人类视觉习惯的兴趣目标的方法。为了进一步提高检测的准确性,该文提出了利用鲁棒前景种子的流形排序进行显著性检测的算法。首先利用角点检测和边缘连接算法得到两个不同的凸包,用它们的交集初步确立目标区域的大致位置;然后利用凸包外边缘作为标准对凸包内的超像素进行相似度检测,将与大部分外边缘相似的超像素去除,得到更准确的目标样本作为前景种子;利用锚点图构建新的图结构表示数据节点之间的关系;接着通过基于前景和背景种子的流形排序算法对图像所有区域进行排序,并得到两种不同的显著性检测图;最后借助代价函数对显著性图进行优化,得到最终的显著性检测结果。经实验表明,与几种经典算法对比,该文方法可以进一步提高显著性算法的精确度和召回率。     

Two-stage salient region detection by exploiting multiple priors

Saliency detection has become a valuable tool for many image processing tasks, like image retargeting, object recognition, and adaptive compression. With the rapid development of the saliency detection methods, people have approved the hypothesis that “the appearance contrast between the salient object and the background is high”, and build their saliency methods on some priors that explain this hypothesis. However, these methods are not satisfactory enough. We propose a two-stage salient region detection method. The input image is first segmented into superpixels. In the first stage, two measures which measure the isolation and distribution of each superpixel are proposed, we consider that both of these two measures are important for finding the salient regions, thus the image-feature-based saliency map is obtained by combining the two measures. Then, in the second stage, we incorporate into the image-feature-based saliency map a location prior map to emphasize the foci of attention. In this algorithm, six priors that explain what is the salient region are exploited. The proposed method is compared with the state-of-the-art saliency detection methods using one of the largest publicly available standard databases, the experimental result indicates that the proposed method has better performance. We also demonstrate how the saliency map of the proposed method can be used to create high quality of initial segmentation masks for subsequent image processing, like Grabcut based salient object segmentation.     

Depth cue enhancement and guidance network for RGB-D salient object detection

Depth maps have been proven profitable to provide supplements for salient object detection in recent years. However, most RGB-D salient object detection approaches ignore that there are usually low-quality depth maps, which will inevitably result in unsatisfactory results. In this paper, we propose a depth cue enhancement and guidance network (DEGNet) for RGB-D salient object detection by exploring the depth quality enhancement and utilizing the depth cue guidance to generate predictions with highlighted objects and suppressed backgrounds. Specifically, a depth cue enhancement module is designed to generate high-quality depth maps by enhancing the contrast between the foreground and the background. Then considering the different characteristics of unimodal RGB and depth features, we use different feature enhancement strategies to strengthen the representation capability of side-output unimodal features. Moreover, we propose a depth-guided feature fusion module to excavate depth cues provided by the depth stream to guide the fusion of multi-modal features by fully making use of different modal properties, thus generating discriminative cross-modal features. Besides, we aggregate cross-modal features at different levels to obtain the final prediction by adopting a pyramid feature shrinking structure. Experimental results on six benchmark datasets demonstrate that the proposed network DEGNet outperforms 17 state-of-the-art methods.     

Study on traffic scene semantic segmentation method based on convolutional neural network

In order to improve the semantic segmentation accuracy of traffic scene,a segmentation method was proposed based on RGB-D image and convolutional neural network.Firstly,on the basis of semi-global stereo matching algorithm,the disparity map was obtained,and the sample library was established by fusing the disparity map D and RGB image into the four-channel RGB-D image.Then,with two different structures,the networks were trained by using two different learning rate adjustment strategy respectively.Finally,the traffic scene semantic segmentation test was carried out with RGB-D image as the input,and the results were compared with the segmentation method based on RGB image.The experimental results show that the proposed traffic scene segmentation algorithm based on RGB-D image can achieve higher semantic segmentation accuracy than that based on RGB image.     

Salient region detection through sparse reconstruction and graph-based ranking

In this paper, we propose a salient region detection algorithm from the point of view of unique and compact representation of individual image. In first step, the original image is segmented into super-pixels. In second step, the sparse representation measure and uniqueness of the features are computed. Then both are ranked on the basis of the background and foreground seeds respectively. Thirdly, a location prior map is used to enhance the foci of attention. We apply the Bayes procedure to integrate computed results to produce smooth and precise saliency map. We compare our proposed algorithm against the state-of-the-art saliency detection methods using four of the largest widely available standard data-bases, experimental results specify that the proposed algorithm outperforms. We also show that how the saliency map of the proposed method is used to discover outline of object, furthermore using this outline our method produce the saliency cut of the desired object.     

Salient object detection via a boundary-guided graph structure

Graph-based salient object detection methods have gained more and more attention recently. However, existing works fail to separate effectively salient object and background in some challenging scenes. Inspired by this observation, we propose an effective salient object detection method based on a novel boundary-guided graph structure. More specifically, the input image is firstly segmented into a series of superpixels. Then we integrate two prior cues to generate the coarse saliency map, a novel weighting mechanism is proposed to balance the proportion of two prior cues according to their performance. Secondly, we propose a novel boundary-guided graph structure to explore deeply the intrinsic relevance between superpixels. Based on the proposed graph structure, an iterative propagation mechanism is constructed to refine the coarse saliency map. Experimental results on four datasets show adequately the superiority of the proposed method than other state-of-the-art methods.     

Saliency detection for panoramic landscape images of outdoor scenes

Saliency detection has been researched for conventional images with standard aspect ratios, however, it is a challenging problem for panoramic images with wide fields of view. In this paper, we propose a saliency detection algorithm for panoramic landscape images of outdoor scenes. We observe that a typical panoramic image includes several homogeneous background regions yielding horizontally elongated distributions, as well as multiple foreground objects with arbitrary locations. We first estimate the background of panoramic images by selecting homogeneous superpixels using geodesic similarity and analyzing their spatial distributions. Then we iteratively refine an initial saliency map derived from background estimation by computing the feature contrast only within local surrounding area whose range and shape are changed adaptively. Experimental results demonstrate that the proposed algorithm detects multiple salient objects faithfully while suppressing the background successfully, and it yields a significantly better performance of panorama saliency detection compared with the recent state-of-the-art techniques.