Saliency detection based on superpixels clustering and stereo disparity

Reliable saliency detection can be used to quickly and effectively locate objects in images. In this paper, a novel algorithm for saliency detection based on superpixels clustering and stereo disparity （SDC） is proposed. Firstly, we use an improved superpixels clustering method to decompose the given image. Then, the disparity of each superpixel is computed by a modified stereo correspondence algorithm. Finally, a new measure which combines stereo disparity with color contrast and spatial coherence is defined to evaluate the saliency of each superpixel. From the experiments we can see that regions with high disparity can get higher saliency value, and the saliency maps have the same resolution with the source images, objects in the map have clear boundaries. Due to the use of superpixel and stereo disparity information, the proposed method is computationally efficient and outperforms some state-of-the-art color- based saliency detection methods.

Object Detection Using SURF and Superpixels

This paper proposes a novel object detection method in which a set of local features inside the superpixels are extracted from the image under analysis acquired by a 3D visual sensor. To increase the segmentation accuracy, the proposed method firstly performs the segmentation of the image, under analysis, using the Simple Linear Iterative Clustering (SLIC) superpixels method. Next the key points inside each superpixel are estimated using the Speed-Up Robust Feature (SURF). These key points are then used to carry out the matching task for every detected keypoints of a scene inside the estimated superpixels. In addition, a probability map is introduced to describe the accuracy of the object detection results. Experimental results show that the proposed approach provides fairly good object detection and confirms the superior performance of proposed scene compared with other recently proposed methods such as the scheme proposed by Mae et al.

High-contrast imaging based on wavefront shaping to improve low signal-to-noise ratio photoacoustic signals using superpixel method

Photoacoustic(PA) imaging has drawn tremendous research interest for various applications in biomedicine and experienced exponential growth over the past decade. Since the scattering effect of biological tissue on ultrasound is two-to three-orders magnitude weaker than that of light, photoacoustic imaging can effectively improve the imaging depth.However, as the depth of imaging further increases, the incident light is seriously affected by scattering that the generated photoacoustic signal is very weak and the signal-to-noise ratio(SNR) is quite low. Low SNR signals can reduce imaging quality and even cause imaging failure. In this paper, we proposed a new wavefront shaping and imaging method of low SNR photoacoustic signal using digital micromirror device(DMD) based superpixel method. We combined the superpixel method with DMD to modulate the phase and amplitude of the incident light, and the genetic algorithm(GA) was used as the wavefront shaping algorithm. The enhancement of the photoacoustic signal reached 10.46. Then we performed scanning imaging by moving the absorber with the translation stage. A clear image with contrast of 8.57 was obtained while imaging with original photoacoustic signals could not be achieved. The proposed method opens new perspectives for imaging with weak photoacoustic signals.

Precise Object Detection Using Iterative Superpixels Grouping Method

The region completeness of object detection is very crucial to video surveillance, such as the pedestrian and vehicle identifications. However, many conventional object detection approaches cannot guarantee the object region completeness because the object detection can be influenced by the illumination variations and clustering backgrounds. In order to overcome this problem, we propose the iterative superpixels grouping （ISPG） method to extract the precise object boundary and generate the object region with high completeness after the object detection. First, by extending the superpixel segmentation method, the proposed ISPG method can improve the inaccurate segmentation problem and guarantee the region completeness on the object regions. Second, the multi- resolution superpixel-based region completeness enhancement method is proposed to extract the object region with high precision and completeness. The simulation results show that the proposed method outperforms the conventional object detection methods in terms of object completeness evaluation.

A Noise-Resistant Superpixel Segmentation Algorithm for Hyperspectral Images

The superpixel segmentation has been widely applied in many computer vision and image process applications.In recent years,amount of superpixel segmentation algorithms have been proposed.However,most of the current algorithms are designed for natural images with little noise corrupted.In order to apply the superpixel algorithms to hyperspectral images which are always seriously polluted by noise,we propose a noiseresistant superpixel segmentation(NRSS)algorithm in this paper.In the proposed NRSS,the spectral signatures are first transformed into frequency domain to enhance the noise robustness;then the two widely spectral similarity measures-spectral angle mapper(SAM)and spectral information divergence(SID)are combined to enhance the discriminability of the spectral similarity;finally,the superpixels are generated with the proposed frequency-based spectral similarity.Both qualitative and quantitative experimental results demonstrate the effectiveness of the proposed superpixel segmentation algorithm when dealing with hyperspectral images with various noise levels.Moreover,the proposed NRSS is compared with the most widely used superpixel segmentation algorithm-simple linear iterative clustering(SLIC),where the comparison results prove the superiority of the proposed superpixel segmentation algorithm.

An review on superpixels generation algorithms

Superpixel segmentation is the oversegmentation of an image into a set of homogeneous regions. Superpixel has many specific properties and has been commonly used as supporting regions for primitives to reduce computations in various computer vision tasks. One property of superpixels is compactness, which is preferred in some applications. In this paper, we give an review on image superpixel segmentation algorithms proposed in recent years. Superpixel segmentation approaches are classified based on the compactness constraint and their main idea are introduced. We also compare these algorithms in visual and evaluate them with five common measurements.

A Research Review of Superpixels Generation Algorithms

Superpixels generation is becoming increasingly popular as a preprocessing in many computer vision applications. A superpixel is an image patch which has uniform pixels intensity and is aligned with intensity edges. Superpixels provide a convenient primitive from which local image features can be computed. So far, there are many methods to generate superpixels. Several main superpixels generation algorithms are summarized in this paper and the advantages and disadvantages of them are analyzed simply. In the end, some applications of superpixels are listed.

基于自适应窗的动态权值代价聚合立体匹配

传统局部匹配算法通常在不同聚合窗口下采用单一权值进行代价聚集,而忽略了不同区域像素点之间的差异性,易导致基于双目视觉测量的立体匹配精度不稳定。为此,提出一种基于自适应窗的动态权值代价聚合立体匹配算法。首先,以梯度信息表征模型为约束条件,构建代价聚合自适应十字交叉窗口。其次,分析视差不连续区域与弱纹理区域的像素特征,建立基于像素距离与色差双阈值权重的代价聚集模型,进而计算各窗口的动态权重影响因子。最后,基于线性迭代聚类法分割视差图区域,剔除奇异的视差值以提高算法匹配精度。实验结果表明,在Middlebury数据集的测试中,所提算法的平均非遮挡区域和全部区域的误匹配率分别为4.11%和5.65%,优于传统匹配算法;4组测量样品的全局长度测量平均相对误差小于1.2%,全局高度测量平均相对误差小于2.7%。实验结果检验了所提算法的有效性和可靠性。

基于核聚类的砂岩图像孔隙分割方法

砂岩孔隙识别是研究孔隙结构的一个重要步骤,采用通用的图像分割算法不易得到理想的图像孔隙分割效果,为此提出了一种使用EfficientNetV2-S模型和核K-Means聚类技术对孔隙进行分割的方法。首先,获得砂岩图像的超像素集合,使用超像素方法预分割输入的致密砂岩图像,构建带标签的孔隙与非孔隙图像库;然后,应用EfficientNetV2-S模型提取砂岩图像的孔隙和非孔隙的语义特征,并结合迁移学习的方法,使用有限的砂岩图像的孔隙和非孔隙样本进行EfficientNetV2-S模型参数学习;最后,设计了一种基于K-Means聚类的区域合并方法——NTK-KCoP方法,根据超像素的语义特征、灰度特征和边缘特征构建目标函数,再由聚类结果合并超像素得到完整的孔隙区域。砂岩CT图像的实验结果验证了所提出的孔隙分割方法的适用性和有效性。

基于先验特征聚类的目标检测优化方法

针对显著目标检测问题在没有任何先验信息的情况下,通过特征聚类和紧致性先验方案实现目标检测优化。优化后的方法包括四个步骤:首先采用超像素预处理将图像分割成超像素,以抑制噪声并降低计算复杂度;其次应用改进的虾群聚类算法对颜色特征进行分类;接着利用二维熵来衡量每个簇的紧密度,并构建背景模型;最后以背景区域与其他区域之间的对比度作为显著特征,并通过设计高斯滤波器增强其显著性。为了更好地评价显著目标检测的精度,本文通过多维评价指标进行优劣性实验分析,实验结果表明,文中算法具有较好的实时性与鲁棒性。

基于多注意力机制与编译图神经网络的高光谱图像分类

针对高光谱图像(Hyperspectral image,HSI)分类研究中小样本学习时,无法达到理想分类效果的问题,以多注意力机制融合、编译图神经网络与卷积神经网络有机结合提出了一种新的高光谱图像分类方法。设计了一种基于混合注意力机制的网络(Multiple mixed attention convolutional neural network,MCNN)与编译图神经网络(Compiled graph neural network,CGNN),在学习样本有限的情况下,其能有效保留HSI的光谱与空间信息。引入的图编码器与图解码器可以有效地映射不规则的HSI地物类别特征信息。设计的多注意力机制可以重点关注一些重要的空间像素特征。研究了不同训练样本下对不同算法学习示例分类的影响,在公共数据集Botswana(BS)的实验表明,本文方法比CEGCN(CNN-enhanced graph convolutional network)、WFCG(Weighted feature fusion of convolutional neural network)算法总体分类精度(Overall classification accuracy,OA)分别高2.72、3.86个百分点。同样在IndianPines(IP)数据集上仅用3%训练样本数据的实验结果显示,本研究方法比CEGCN与WFCG算法的OA分别高0.44、1.42个百分点。说明本研究提出的方法不仅对HSI具有良好的空间与光谱信息感知能力,而且在微小学习数据下仍然表现出强有力的分类准确性。

基于皮尔森相关系数的自适应SLIC超像素分割算法

针对简单线性迭代聚类(Simple Linear Iterative Clustering,SLIC)算法对不同图像自适应性差的问题,提出了一种基于皮尔森相关系数的自适应SLIC超像素图像分割算法。首先,通过量化非间隔进行图像预处理,并计算颜色熵作为图像复杂度,从而确定所需分割的超像素个数。其次,利用皮尔森相关系数作为相似性度量函数。最后,通过纹理特征对类内异常点进行滤除,确保种子点更新的准确性。实验结果表明,在超像素个数相同的情况下,基于皮尔森相关系数的自适应SLIC超像素图像分割算法相比主流超像素分割算法,可以获得更高的边缘命中率以及更低的欠分割率,性能优于LSC(Linear Spectral Clustering)、SLIC和SLIC0(Simple Linear Iterative Clustering Zero)算法。

基于改进超像素采样的立体匹配网络

针对立体匹配中细节丢失、有遮挡,以及无纹理区域匹配精度低的问题,提出了一种基于改进超像素采样的立体匹配方法。首先,利用改进的超像素采样方法对用于立体匹配的高分辨率输入图像进行下采样,随后,将下采样后的图像对输入到立体匹配网络中,利用权值共享的卷积网络进行特征提取,使用3D卷积获取特征融合后的Cost Volume并生成视差图,再将输出的视差图进行上采样还原为最终的视差图。针对超像素采样过程中容易丢失细节从而影响后续立体匹配精度的问题,引入特征金字塔注意力模块(Feature Pyramid Attention,FPA)和改进的残差结构。根据上述两个方面的创新,提出了基于超像素采样的立体匹配网络FPSMnet(Feature Pyramid Stereo Matching Network),并选取、划分图像数据集BSDS500和NYUv2作为超像素采样的训练、验证和测试的数据集。立体匹配实验结果表明,与基准方法相比,所提算法在SceneFlow和HR-VS数据集上的平均像素误差分别下降了0.25和0.52,在不影响运行时间的前提下提高了匹配精度。

基于解耦区域校准的高分辨率超像素生成算法

超像素分割是计算机视觉领域的一项重要任务,该任务将具有相似属性的像素分组到称为超像素的簇中.图像超像素不仅可以增益图像注释,而且还是各种下游应用的基础,如分割、光流估计和深度估计.尽管超像素分割技术取得了显著进展,特别是随着深度学习方法的出现,但现有解决方案由于GPU内存和计算能力的限制,一直无法有效处理高分辨率图像.针对这个问题,作者提出了一种名为区域解耦校准的高分辨率超像素网络(Patch Calibration Network,PCNet)的新型深度学习框架,通过采用解耦的一致性学习策略,解决了现有方法的局限性.这种方法允许通过从低分辨率输入预测高分辨率输出来高效生成高分辨率超像素结果,从而绕过了GPU内存限制.PCNet的一个关键贡献是解耦的区域块校准(DPC)分支,它将高分辨率图像块作为额外输入,以保留细节并增强边界像素分配.为了改善边界像素的识别,作者利用二进制掩模设计了一种动态引导训练机制.这种机制鼓励网络专注于区域内的主要边界,将任务从多类分类简化为二分类问题.这一创新策略不仅减少了网络优化的复杂性,而且显著提高了边界检测的精度.本文通过在包括Mapillary Vistas、BIG和新创建的Face-Human数据集在内的多样化数据集上进行广泛的实验,证明了PCNet的有效性.结果表明,PCNet能够成功处理5K分辨率图像,并与现有的最先进的SCN方法相比,实现了更优越的性能,后者在处理高分辨率输入时存在困难.作者的贡献包括开发了PCNet,一种针对高分辨率超像素分割的深度学习解决方案,引入了解耦的区域校准架构,并构建了一个超高分辨率基准测试集,用于评估高分辨率场景中超像素分割算法的性能.本文首先回顾了超像素分割领域的相关工作,然后详细介绍了PCNet框架,接着展示了实验结果并与最先进的方法进行了比较.结论部分总结了研究结果并概述了未来研究的潜在方向.代码、预训练模型和新的基准数据集的可用性无疑将促进高分辨率超像素分割领域的进一步发展.总之,本文在超像素分割领域提供了一个重要的进步,提供了一种能够高效、准确处理高分辨率图像的解决方案.所提出的PCNet框架,凭借其创新的DPC分支和动态引导训练机制,为未来在计算机视觉领域的研究和应用提供了一个有前景的方向.本文的代码、预训练模型以及新构建的评估基准数据集可在https://github.com/wangyxxjtu/PCNet上获取.

BYOL框架下的自监督高光谱图像分类

高光谱图像可以获取波段连续的图谱合一的立体数据,其具有丰富的图谱信息,能区分不同物质的类别,被广泛应用于各种遥感勘测领域。但在实际中高光谱图像的标注需要耗费大量的人力、财力和时间,可用的标注样本数量较少,难以通过训练来获得准确的分类结果,所以针对于只有少量标记样本的高光谱图像分类是一个挑战。近年来,自监督学习(Self-supervised Learning,SSL)已成为一种有效的方法,可以减少高光谱图像分类对昂贵的数据标注的依赖。SSL方法通过学习在同一图像的不同视图之间产生的潜在特征,在自然图像分类中取得了较高的分类精度。为了探索SSL方法在高光谱图像分类中的潜力,一种Bootstrap Your Own Latent(BYOL)框架下的自监督高光谱图像分类方法(BSSL)被提出。该方法通过引用自监督的图像特征学习框架BYOL,可以不需要负样本对,利用空间光谱相似的同类样本对进行网络训练及参数微调,提取到更具判别性特征。具体来说,该方法主要包括四个部分:BYOL的预训练、超像素聚类、基于“相似对”的BYOL的再训练和最终分类。为了验证该方法的有效性,在三个公开数据集上进行测试,并与五种先进的无监督、自监督分类方法SuperPCA、S3PCA、ContrastNet、SSCL和N2SSL进行对比,在Indian Pines和Salinas数据集上,BSSL方法的总体分类精度(OA)、平均分类精度(AA)、Kappa系数、召回率(recall)和f1分数(f1-score)都取得了更优值。其中在Indian Pines数据集上,OA分别比SuperPCA,S3PCA,ContrastNet,SSCL和N2SSL提高了1.32%,1.05%,5.68%,3.12%和1.27%。而在University of Pavia数据集上,BSSL方法表现没有那么出色,但在综合分类性能上也表现最优。这表明BSSL方法更适用于地物区域面积较大且分布较集中的场景,因为这对于超像素聚类来说更友好。

基于天空检测和超像素分割的图像去雾方法

针对经典图像去雾算法在边缘区域易产生光晕效应、天空等明亮区域还原失真、色调偏移等问题,提出一种基于天空检测和超像素分割的改进暗通道图像去雾新方法(Dark Channel Prior based on Sky Detection and Super Pixel,SSPDCP).首先对雾图采用HSV变换提取亮度分量进行自适应阈值分割;然后应用图像连通分析技术识别天空域;接着利用天空域估计大气光值,针对天空和非天空区域分别建立各自的透射率计算模型,并基于构建的超像素级透射率融合模型获得融合透射率图,以促进边界区域的平滑过渡,采用多尺度引导滤波精化透射率图;最后应用大气散射模型完成图像复原并进行亮度增强处理,实现无雾图像的自然恢复.该方法识别的天空区域较为连续完整,以超像素代替方形窗口可以有效克服局部块效应的影响,大气光值和透射率图估计更为客观准确.从主观定性和客观定量评价方面来看,该方法复原的图像具有整体误差小、信噪比优良、结构相似度高等优势.本文所提出的图像去雾新方法能有效抑制边缘区域的光晕效应,且复原的天空区域明亮自然,图像去雾质量相比现有方法有进一步提升.

一种小样本情境下的高光谱图像分类算法

Gabor滤波器是一种常见的空间特征提取技术,在针对高光谱图像分类中已标记样本稀缺的问题上,该算法通过设置不同方向的多个3D-Gabor滤波器,生成大量多视图。在多视图数据基础上生成多个图连接实现标签传播,将多个图标签传播后的分类结果融合得到预测标结果。而超像素主成分分析法算法则是一种简单但非常有效的无监督特征提取方法,将预测结果与加入了超像素主成分分析法的分类器相加权融合得到更为准确的分类结果。将算法在3个数据集上进行仿真实验,结果表明通过应用Gabor滤波器的传统高光谱图像分类算法存在运算量大且耗时长,而该算法能够在保证精度的同时有效减少计算及时间上的消耗,节约成本。

基于假设检验的SAR图像水域分割增强方法

水域图像分割常应用于洪灾监测、地表水资源勘察等工作。合成孔径雷达(Synthetic Aperture Radar,SAR)图像常常带有阴影和干扰斑,导致现有的水域分割结果容易出现水域连通性差、水域边缘粗糙等问题。为了更好地利用目标的边缘信息解决这些问题,分别独立采用语义分割和超像素分割得到水域预测结果与超像素聚类掩码,利用假设检验模型融合2种方法的分割结果。在U-Net、SegNet、DeepLabV3+和超像素图卷积网络(Superpixel Graph Convolutional Network,SGCN)4种语义分割方法进行实验,分割精度均有提升。在U-Net上提升最大,平均交并比(mean Intersection over Union,mIoU)由78.24%提升至83.39%。对比了3种不同的超像素分割方法的改善效果,实验结果表明,所提方法在线性谱聚类(Linear Spectral Clustering,LSC)中表现最佳。

基于插值超分辨的双目三维重建方法

基于双目立体匹配重建物体表面三维形貌时,其匹配精度往往受限于传感器尺寸、镜头焦距和光源环境等物理条件。针对此问题,本文提出了一种基于插值超分辨的双目表面三维重建方法。首先,在图像预处理阶段,建立基于小波变换与双直方图均衡融合的图像增强方法,克服传统双目视觉受限于复杂环境光干扰等问题;其次,构建基于拉格朗日与三次多项式插值的超分辨算法,提升图像像素密度,为双目匹配代价计算阶段增加图像细节,从而提高匹配精度;最后,基于SLIC算法对目标图像进行分割,并针对各分割区域分别做二次曲面拟合,进而获得与物体实际表面更为贴合的高度曲线,从而降低重建误差并可提高重建精度。实验结果表明:5组测量样品的全局高度测量平均相对误差为±2.3%,实验平均测量时长为1.8828 s,最大时长为1.9362 s,较传统方法有明显提升。实验分析结果验证了本文所提方法的有效性。

基于超像素分割的图注意力网络的高光谱图像分类

针对卷积神经网络(CNN)仅能应用于欧氏数据,无法有效获取像素间的全局关系特征以及长距离上下文信息的问题,构建一个基于超像素分割的图注意力网络SSGAT.该网络将超像素分割后的超像素块视为图结构中的图节点,有效减少了图结构的复杂度,并降低了分类图的噪声.在3个数据集上对SSGAT及对比算法的分类精度进行测试,分别获得了94.11%,95.22%,96.37%的总体分类精度.结果表明该方法性能优异,在处理大尺度区域的分类问题时优势明显.