Master-CAM: Multi-scale fusion guided by Master map for high-quality class activation maps

Class Activation Map (CAM) is one of the most popular approaches to visually explain the convolutional neural networks (CNNs). To obtain fine-grained saliency maps, some works fuse saliency signals of the same image at larger scales. However, existing methods based on multi-scale fusion cannot effectively remove the noise from larger-scale images. In this paper, we propose Master-CAM, which uses Master map to guide multi-scale fusion process to obtain a high-quality class activation map. Master-CAM utilizes the general localization ability of the Master map to reduce the noise of the maps. We call the one with the general localization ability among the saliency maps from the same image as Master map, which is the saliency map of the original-scale input in the multi-scale scenario. In addition, we also present a simple yet effective fusion strategy, Master-Fusion, which is derived from the fusion operation in Master-CAM. Master-Fusion strategy can be easily attached to some saliency methods to improve the performance of these methods. We show through qualitative and quantitative experiments that the proposed Master-CAM outperforms the state-of-the-art methods in different CNN frameworks and datasets.     

融合深度模型和传统模型的显著性检测

目的 显著性检测是图像和视觉领域一个基础问题,传统模型对于显著性物体的边界保留较好,但是对显著性目标的自信度不够高,召回率低,而深度学习模型对于显著性物体的自信度高,但是其结果边界粗糙,准确率较低。针对这两种模型各自的优缺点,提出一种显著性模型以综合利用两种方法的优点并抑制各自的不足。方法 首先改进最新的密集卷积网络,训练了一个基于该网络的全卷积网络（FCN）显著性模型,同时选取一个现有的基于超像素的显著性回归模型,在得到两种模型的显著性结果图后,提出一种融合算法,融合两种方法的结果以得到最终优化结果,该算法通过显著性结果Hadamard积和像素间显著性值的一对一非线性映射,将FCN结果与传统模型的结果相融合。结果 实验在4个数据集上与最新的10种方法进行了比较,在HKU-IS数据集中,相比于性能第2的模型,F值提高了2.6%;在MSRA数据集中,相比于性能第2的模型,F值提高了2.2%,MAE降低了5.6%;在DUT-OMRON数据集中,相比于性能第2的模型,F值提高了5.6%,MAE降低了17.4%。同时也在MSRA数据集中进行了对比实验以验证融合算法的有效性,对比实验结果表明提出的融合算法改善了显著性检测的效果。结论 本文所提出的显著性模型,综合了传统模型和深度学习模型的优点,使显著性检测结果更加准确。     

超像素条件随机场下的RGB-D视频显著性检测

目的 视觉显著性在众多视觉驱动的应用中具有重要作用,这些应用领域出现了从2维视觉到3维视觉的转换,从而基于RGB-D数据的显著性模型引起了广泛关注。与2维图像的显著性不同,RGB-D显著性包含了许多不同模态的线索。多模态线索之间存在互补和竞争关系,如何有效地利用和融合这些线索仍是一个挑战。传统的融合模型很难充分利用多模态线索之间的优势,因此研究了RGB-D显著性形成过程中多模态线索融合的问题。方法 提出了一种基于超像素下条件随机场的RGB-D显著性检测模型。提取不同模态的显著性线索,包括平面线索、深度线索和运动线索等。以超像素为单位建立条件随机场模型,联合多模态线索的影响和图像邻域显著值平滑约束,设计了一个全局能量函数作为模型的优化目标,刻画了多模态线索之间的相互作用机制。其中,多模态线索在能量函数中的权重因子由卷积神经网络学习得到。结果 实验在两个公开的RGB-D视频显著性数据集上与6种显著性检测方法进行了比较,所提模型在所有相关数据集和评价指标上都优于当前最先进的模型。相比于第2高的指标,所提模型的AUC（area under curve）,sAUC（shuffled AUC）,SIM（similarity）,PCC（Pearson correlation coefficient）和NSS（normalized scanpath saliency）指标在IRCCyN数据集上分别提升了2.3%,2.3%,18.9%,21.6%和56.2%;在DML-iTrack-3D数据集上分别提升了2.0%,1.4%,29.1%,10.6%,23.3%。此外还进行了模型内部的比较,验证了所提融合方法优于其他传统融合方法。结论 本文提出的RGB-D显著性检测模型中的条件随机场和卷积神经网络充分利用了不同模态线索的优势,将它们有效融合,提升了显著性检测模型的性能,能在视觉驱动的应用领域发挥一定作用。     

自适应融合局部和全局稀疏表示的图像显著性检测

针对基于局部或全局稀疏表示的图像显著性检测方法频繁出现提取对象不完整、边界不光滑及噪声消除不干净等问题,提出自适应融合局部和全局稀疏表示的图像显著性检测方法。首先,对原始图像进行分块处理,利用图像块代替像素操作,降低算法复杂度;其次,对分块后的图像进行局部稀疏表示,即：针对每一个图像块,选取其周围的若干图像块生成过完备字典,基于该字典对图像块进行稀疏重构,得到原始图像的初始局部显著图,该显著图能够有效提取显著性目标的边缘信息;接着,对分块后的图像进行全局稀疏表示,与局部稀疏表示过程类似,不同的是针对每一个图像块所生成的字典来源于图像四周边界处的图像块,这样可以得到能有效检测出显著性目标内部区域的初始全局显著图;最后,将初始局部和全局显著图进行自适应融合,生成最终显著图。实验结果表明,提出算法在查准率（precision）、查全率（recall）及F-measure等指标上优于几种经典的图像显著性检测方法。     

Saliency detection via image sparse representation and color features combination

Saliency detection is a technique to analyze image surroundings to extract relevant regions from the background. In this paper, we propose a simple and effective saliency detection method based on image sparse representation and color features combination. First, the input image is segmented into non-overlapping super-pixels, so as to perform the saliency detection at the region level to reduce computational complexity. Then, a background optimization selection scheme is used to construct an appropriate background template. Based on this, a primary saliency map is obtained by using image sparse representation. Next, through the linear combination of color coefficients we generate an improved saliency map with more prominent salient regions. Finally, the two saliency maps are integrated within Bayesian framework to obtain the final saliency map. Experimental results show that the proposed method has desirable detection performance in terms of detection accuracy and running time.

     

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...     

基于视觉显著性图与似物性的对象检测

为了能够准确地检测出图像中的显著性对象,提出了一种新的基于视觉显著性图与似物性的对象检测算法。该算法首先在图像上提取大量具有较高似物性度量的矩形窗口,并估算出对象可能出现的位置,将窗口级的似物性度量转换到像素级的似物性度量;然后把原始显著性图与像素级的似物性图进行融合,生成加权显著性图,分别二值化原始显著性图和加权显著性图,利用凸包检测得到最大查找窗口区域与种子窗口区域;最后结合边缘概率密度搜索出最优的对象窗口。在公开数据集MSRA-B上的实验结果表明,该算法在准确率、召回率以及F-测度方面优于最大化显著区域检测算法、区域密度最大化算法以及似物性对象检测算法等已有的多种算法。     

一种随机的视觉显著性检测算法

图像处理与模式识别技术一样,依赖于高质量的视觉显著性图(saliency map)才能得到较好的处理结果.现有的视觉显著性检测技术通常只能检测得到粗糙的视觉显著性图;这些粗糙的视觉显著性图应用于图像处理中将严重影响图像处理的最终结果.本文提出了一种随机的基于内容的视觉显著性区域检测算法;该算法整合多层次粗糙的视觉显著性图到结果显著性图中,并逐步自适应地精化可信度不高的显著性值,最终得到一个考虑了多尺度特征的精细的视觉显著性结果.因为随机算法具有执行效率高,占用内存少等特点;本文的高效随机视觉显著性检测算法不需要建立额外的辅助数据结构来加速算法,只需占用少量内存就能快速检测出精细的高质量视觉显著性结果.并且高效随机的视觉显著性检测算法可以直接移植到GPU上并行执行;大量的实验结果表明本文的算法可以得到更加精细的显著性结果,这些精细的显著性结果应用于基于内容的图像缩放中得到了较好的处理结果.     

显著性引导及不确定性监督的深度编解码网络

基于U-Net的编码-解码网络及其变体网络在医学图像语义分割任务中取得了卓越的分割性能.然而,网络在特征提取过程中丢失了部分空间细节信息,影响了分割精度.另一方面,在多模态的医学图像语义分割任务中,这些模型的泛化能力和鲁棒性不理想.针对以上问题,本文提出一种显著性引导及不确定性监督的深度卷积编解码网络,以解决多模态医学图像语义分割问题.该算法将初始生成的显著图和不确定概率图作为监督信息来优化语义分割网络的参数.首先,通过显著性检测网络生成显著图,初步定位图像中的目标区域;然后,根据显著图计算不确定分类的像素点集合,生成不确定概率图;最后,将显著图和不确定概率图与原图像一同送入多尺度特征融合网络,引导网络关注目标区域特征的学习,同时增强网络对不确定分类区域和复杂边界的表征能力,以提升网络的分割性能.实验结果表明,本文算法能够捕获更多的语义信息,在多模态医学图像语义分割任务中优于其他的语义分割算法,并具有较好的泛化能力和鲁棒性.     

融合双目多维感知特征的立体视频显著性检测

目的 立体视频能提供身临其境的逼真感而越来越受到人们的喜爱,而视觉显著性检测可以自动预测、定位和挖掘重要视觉信息,可以帮助机器对海量多媒体信息进行有效筛选。为了提高立体视频中的显著区域检测性能,提出了一种融合双目多维感知特性的立体视频显著性检测模型。方法 从立体视频的空域、深度以及时域3个不同维度出发进行显著性计算。首先,基于图像的空间特征利用贝叶斯模型计算2D图像显著图;接着,根据双目感知特征获取立体视频图像的深度显著图;然后,利用Lucas-Kanade光流法计算帧间局部区域的运动特征,获取时域显著图;最后,将3种不同维度的显著图采用一种基于全局-区域差异度大小的融合方法进行相互融合,获得最终的立体视频显著区域分布模型。结果 在不同类型的立体视频序列中的实验结果表明,本文模型获得了80%的准确率和72%的召回率,且保持了相对较低的计算复杂度,优于现有的显著性检测模型。结论 本文的显著性检测模型能有效地获取立体视频中的显著区域,可应用于立体视频/图像编码、立体视频/图像质量评价等领域。     

Saliency detection based on self-adaptive multiple feature fusion for remote sensing images

ABSTRACT

Saliency detection has been revealed an effective and reliable approach to extract the region of interest (ROI) in remote sensing images. However, most existing saliency detection methods employing multiple saliency cues ignore the intrinsic relationship between different cues and do not distinguish the diverse contributions of different cues to the final saliency map. In this paper, we propose a novel self-adaptively multiple feature fusion model for saliency detection in remote sensing images to take advantage of this relationship to improve the accuracy of ROI extraction. First, we take multiple feature channels, namely colour, intensity, texture and global contrast into consideration to produce primary feature maps. Particularly, we design a novel method based on dual-tree complex wavelet transform for remote sensing images to generate texture feature pyramids. Then, we introduce a novel self-adaptive multiple feature fusion method based on low-rank matrix recovery, in which the significances of feature maps are ranked by the low rank constraint recovery, and subsequently multiple features’ contributions are allocated adaptively to produce the final saliency map. Experimental results demonstrate that our proposal outperforms the state-of-the-art methods.     

结合语义辅助和边缘特征的显著对象检测

目的 现有的显著对象检测模型能够很好地定位显著对象,但是在获得完整均匀的对象和保留清晰边缘的任务上存在不足。为了得到整体均匀和边缘清晰的显著对象,本文提出了结合语义辅助和边缘特征的显著对象检测模型。方法 模型利用设计的语义辅助特征融合模块优化骨干网的侧向输出特征,每层特征通过语义辅助选择性融合相邻的低层特征,获得足够的结构信息并增强显著区域的特征强度,进而检测出整体均匀的显著对象。通过设计的边缘分支网络以及显著对象特征得到精确的边缘特征,将边缘特征融合到显著对象特征中,加强特征中显著对象边缘区域的可区分性,以便检测出清晰的边缘。同时,本文设计了一个双向多尺度模块来提取网络中的多尺度信息。结果 在4种常用的数据集ECSSD （extended complex scene saliency dataset）、DUT-O （Dalian University of Technology and OMRON Corporation）、HKU-IS和DUTS上与12种较流行的显著模型进行比较,本文模型的最大F值度量（max F-measure,MaxF）和平均绝对误差（mean absolution error,MAE）分别是0.940、0.795、0.929、0.870和0.041、0.057、0.034、0.043。从实验结果看,本文方法得到的显著图更接近真值图,在MaxF和MAE上取得最佳性能的次数多于其他12种方法。结论 本文提出的结合语义辅助和边缘特征的显著对象检测模型十分有效。语义辅助特征融合和边缘特征的引入使检测出的显著对象更为完整均匀,对象的边缘区分性也更强,多尺度特征提取进一步改善了显著对象的检测效果。     

Capture and fusion of 3d surface texture

Image fusion is a process that multiple images of a scene are combined to form a single image. The aim of image fusion is to preserve the full content and retain important features of each original image. In this paper, we propose a novel approach based on wavelet transform to capture and fusion of real-world rough surface textures, which are commonly used in multimedia applications and referred to as3D surface texture. These textures are different from 2D textures as their appearances can vary dramatically with different illumination conditions due to complex surface geometry and reflectance properties. In our approach, we first extract gradient/height and albedo maps from sample 3D surface texture images as their representation. Then we measure saliency of wavelet coefficients of these 3D surface texture representations. The saliency values reflect the meaningful content of the wavelet coefficients and are consistent with human visual perception. Finally we fuse the gradient/height and albedo maps based on the measured saliency values. This novel scheme aims to preserve the original texture patterns together with geometry and reflectance characteristics from input images. Experimental results show that the proposed approach can not only capture and fuse 3D surface texture under arbitrary illumination directions, but also has the ability to retain the surface geometry properties and preserve perceptual features in the original images.     

基于频域多尺度分析的显著性检测

针对现有显著性检测方法得到的显著区域不完整以及缺乏生物学依据的不足,提出一种基于频域多尺度分析的图像显著性检测方法.首先利用小波变换将输入图像的离散余弦变换(DCT)系数的幅度谱进行多尺度分解,计算得到多尺度下的空间域视觉显著图,然后依据显著性评价函数选出较优显著图,最后以自适应权重合成输入场景的视觉显著图.对不同类型数据集进行实验,包括心理物理学模板数据集、人眼注视轨迹数据集及显著目标分割数据集(包括ASD和ECSSD数据集),该方法对于多类型数据集在P-R曲线、ROC曲线及AUC指标等客观评价标准上均取得较高精确度,且在计算速度统计中计算较快,表明该方法优于其他经典的显著性检测方法.     

Informative patches sampling for image classification by utilizing bottom-up and top-down information

In image classification based on bag of visual words framework, image patches used for creating image representations affect the classification performance significantly. However, currently, patches are sampled mainly based on processing low-level image information or just extracted regularly or randomly. These methods are not effective, because patches extracted through these approaches are not necessarily discriminative for image categorization. In this paper, we propose to utilize both bottom-up information through processing low-level image information and top-down information through exploring statistical properties of training image grids to extract image patches. In the proposed work, an input image is divided into regular grids, each of which is evaluated based on its bottom-up information and/or top-down information. Subsequently, every grid is assigned a saliency value based on its evaluation result, so that a saliency map can be created for the image. Finally, patch sampling from the input image is performed on the basis of the obtained saliency map. Furthermore, we propose a method to fuse these two kinds of information. The proposed methods are evaluated on both object categories and scene categories. Experiment results demonstrate their effectiveness.     

A novel edge-oriented framework for saliency detection enhancement

Mixed visual scenes and cluttered background commonly exist in natural images, which forms a challenge for saliency detection. In dealing with complex images, there are two kinds of deficiencies in the existing saliency detection methods: ambiguous object boundaries and fragmented salient regions. To address these two limitations, we propose a novel edge-oriented framework to improve the performance of existing salient detection methods. Our framework is based on two interesting insights: 1) human eyes are sensitive to the edges between foreground and background even there is hardly any difference in terms of saliency, 2) Guided by semantic integrity, human eyes tend to view a visual scene as several objects, rather than pixels or superpixels. The proposed framework consists of the following three parts. First, an edge probability map is extracted from an input image. Second, the edge-based over-segmentation is obtained by sharpening the edge probability map, which is ultilized to generate edge-regions using an edge-strength based hierarchical merge model. Finally, based on the prior saliency map generated by existing methods, the framework assigns each edge-region with a saliency value. Based on four publically available datasets, the experiments demonstrate that the proposed framework can significantly improve the detection results of existing saliency detection models, which is also superior to other state-of-the-art methods.     

结合布尔图和灰度稀缺性的小目标显著性检测

目的 基于超像素分割的显著物体检测模型在很多公开数据集上表现优异,但在实际场景应用时,超像素分割的数量和大小难以自适应图像和目标大小的变化,从而使性能下降,且分割过多会耗时过大。为解决这一问题,本文提出基于布尔图和灰度稀缺性的小目标显著性检测方法。方法 利用布尔图的思想,提取图像中较为突出的闭合区域,根据闭合区域的大小赋予其显著值,形成一幅显著图;利用灰度稀缺性,为图像中的稀缺灰度值赋予高显著值,抑制烟雾、云、光照渐晕等渐变背景,生成另一幅显著图;将两幅显著图融合,得到具有全分辨率、目标突出且轮廓清晰的显著图。结果 在3个数据集上与14种显著性模型进行对比,本文算法生成的显著图能有效抑制背景,并检测出多个小目标。其中,在复杂背景数据集上,本文算法具有最高的F值（F-measure）和最小的MAE（mean absolute error）值,AUC（area under ROC curve）值仅次于DRFI（discriminative regional feature integration）和ASNet（attentive saliency network）模型,AUC和F-measure值比BMS（Boolean map based saliency）模型分别提高了1.9%和6.9%,MAE值降低了1.8%;在SO200数据集上,本文算法的F-measure值最高,MAE值仅次于ASNet,F-measure值比BMS模型提高了3.8%,MAE值降低了2%;在SED2数据集上,本文算法也优于6种传统模型。在运行时间方面,本文算法具有明显优势,处理400×300像素的图像时,帧频可达12帧/s。结论 本文算法具有良好的适应性和鲁棒性,对于复杂背景下的小目标具有良好的显著性检测效果。     

基于视觉显著性的水面垃圾目标检测

针对实际水面复杂环境提出了一种基于视觉显著性的水面垃圾目标检测算法。首先对输入图像进行超像素分割,在CIELab、RGB和HSV颜色空间中提取超像素级的显著性特征,然后使用随机森林回归器将显著性特征进行融合得到疑似显著性图,并使用自适应阈值分割得到疑似二值显著性图,最后使用MLP分类器对原始图像中的疑似垃圾目标区域进行判别,去除水波、倒影和反光的干扰,最终检测出水面的垃圾目标。实验结果表明所提基于视觉显著性的水面垃圾目标检测算法的性能优于其他水面目标检测算法。     

多先验特征与综合对比度的图像显著性检测

目的 图像的显著性检测在计算机视觉中应用非常广泛,现有的方法通常在复杂背景区域下表现不佳,由于显著性检测的低层特征并不可靠,同时单一的特征也很难得到高质量的显著图。提出了一种通过增加特征的多样性来实现显著性检测的方法。方法 在高层先验知识的基础上,对背景先验特征和中心先验特征重新进行了定义,并考虑人眼视觉一般会对暖色调更为关注,从而加入颜色先验。另外在图像低层特征上使用目前较为流行的全局对比度和局部对比度特征,在特征融合时针对不同情况分别采取线性和非线性的一种新的融合策略,得到高质量的显著图。结果 在MSRA-1000和DUT-OMRON两个公开数据库进行对比验证,实验结果表明,基于多先验特征与综合对比度的图像显著性检测算法具有较高的查准率、召回率和F-measure值,相较于RBD算法均提高了1.5%以上,综合性能均优于目前的10种主流算法。结论 相较于基于低层特征和单一先验特征的算法,本文算法充分利用了图像信息,能在突出全局对比度的同时也保留较多的局部信息,达到均匀突出显著性区域的效果,有效地抑制复杂的背景区域,得到更加符合视觉感知的显著图。     

基于低秩矩阵恢复的视觉显著性目标检测与细化

基于低秩矩阵恢复(low-rank matrix recovery,LRMR)的显著性目标检测模型将图像特征分解为与背景关联的低秩分量和与显著性目标相关联的稀疏分量,并从稀疏分量中获得显著性目标.现有的显著性检测方法很少考虑低秩分量与稀疏分量之间的相互关系,导致检测的显著性目标零散或不完整.为此,提出基于低秩矩阵恢复的显著性目标检测与细化方法来规避该限制.首先,所提方法采用ell_1范数稀疏约束和拉普拉斯正则项对初始显著图进行计算;在显著性细化阶段,由于非局部的ell_0优化可以有效地对显著性区域及其邻接区域之间的相互关系进行建模,结合初始显著图,采用非局部ell_0梯度优化,最小化显著性区域中显著值的变化,从而保证显著性目标的完整性.在4个显著性目标检测数据集上进行实验,通过实验结果验证所提算法的优越性.