The Use of Boolean Model for Texture Analysis of Grey Images

We generalize here the use of the 1D Boolean model for the analysis of grey level textures. Each grey image is first split into eight binary images using different criteria. Each of these binary images is separately analysed with the help of the 1D Boolean model and features are extracted from it. The final grey texture recognition is performed on the basis of these features using several classification criteria. Experiments have been carried out using an image database of 30 grey level textures, all of them with 512×512 pixels in size, obtaining correct classification rates between 95% and 100%, according to the classification criterion used.     

Sub-pattern texture recognition using intelligent focal-plane imaging sensor of small window-size

In this paper we demonstrate how to use statistical evaluation for texture recognition in the case of window-size of the imaging focal-plane sensor being smaller than the pattern of the texture. The evaluation method is similar to the sub-pixel pattern recognition developed by the first author. We have reported in an earlier publication on the development of a new single-chip texture classifier smart-sensor system, whose main part is a cellular nonlinear network (CNN) VLSI chip. This architecture is very fast but it has a limited window-size. Now we show that this architecture can effectively recognize textures of periodicity larger than the window-size. As a result, we recognized 15 Brodatz-textures by using a 20 × 22 CNN chip with a 0.4% error-rate.     

残差密集空间金字塔网络的城市遥感图像分割

目的 遥感图像语义分割是根据土地覆盖类型对图像中每个像素进行分类,是遥感图像处理领域的一个重要研究方向。由于遥感图像包含的地物尺度差别大、地物边界复杂等原因,准确提取遥感图像特征具有一定难度,使得精确分割遥感图像比较困难。卷积神经网络因其自主分层提取图像特征的特点逐步成为图像处理领域的主流算法,本文将基于残差密集空间金字塔的卷积神经网络应用于城市地区遥感图像分割,以提升高分辨率城市地区遥感影像语义分割的精度。方法 模型将带孔卷积引入残差网络,代替网络中的下采样操作,在扩大特征图感受野的同时能够保持特征图尺寸不变;模型基于密集连接机制级联空间金字塔结构各分支,每个分支的输出都有更加密集的感受野信息;模型利用跳线连接跨层融合网络特征,结合网络中的高层语义特征和低层纹理特征恢复空间信息。结果 基于ISPRS （International Society for Photogrammetry and Remote Sensing） Vaihingen地区遥感数据集展开充分的实验研究,实验结果表明,本文模型在6种不同的地物分类上的平均交并比和平均F₁值分别达到69.88%和81.39%,性能在数学指标和视觉效果上均优于SegNet、pix2pix、Res-shuffling-Net以及SDFCN （symmetrical dense-shortcut fully convolutional network）算法。结论 将密集连接改进空间金字塔池化网络应用于高分辨率遥感图像语义分割,该模型利用了遥感图像不同尺度下的特征、高层语义信息和低层纹理信息,有效提升了城市地区遥感图像分割精度。     

多通道卷积的图像超分辨率方法

目的 超分辨率技术在实际生活中具有较为广泛的应用。经典的基于卷积神经网络的超分辨率（SRCNN）方法存在重建图像纹理结构模糊以及网络模型训练收敛过慢等问题。针对这两个问题,在SRCNN的基础上,提出一种多通道卷积的图像超分辨率（MCSR）算法。方法 通过增加残差链接,选择MSRA初始化方法对网络权值进行初始化,加快模型收敛;引入多通道映射提取更加丰富的特征,使用多层3×3等小卷积核代替单层9×9等大卷积核,更加有效地利用特征,增强模型的超分辨率重构效果。结果 MCSR迭代4×10⁶次即可收敛,在Set5与Set14数据集上边长放大3倍后的平均峰值信噪比分别是32.84 dB和29.28 dB,与SRCNN相比提升显著。结论 MCSR收敛速度更快,并且可以生成轮廓清晰的高分辨率图像,超分辨率效果更加优秀。     

Make U-Net Greater: An Easy-to-Embed Approach to Improve Segmentation Performance Using Hypergraph

Cardiac anatomy segmentation is essential for cardiomyopathy clinical diagnosis and treatment planning. Thus, accurate delineation of target volumes at risk in cardiac anatomy is important. However, manual delineation is a time-consuming and labor-intensive process for cardiologists and has been shown to lead to significant inter-and intra-practitioner variability. Thus, computer-aided or fully automatic segmentation methods are required. They can significantly economize on manpower and improve treatment efficiency. Recently, deep convolutional neural network (CNN) based methods have achieved remarkable successes in various kinds of vision tasks, such as classification, segmentation and object detection. Semantic segmentation can be considered as a pixel-wise task, it requires high-level abstract semantics information while maintaining spatial detail contexts. Long-range context information plays a crucial role in this scenario. However, the traditional convolution kernel only provides the local and small size of the receptive field. To address the problem, we propose a plug-and-play module aggregating both local and global information (aka LGIA module) to capture the high-order relationship between nodes that are far apart. We incorporate both local and global correlations into hypergraph which is able to capture high-order relationships between nodes via the concept of a hyperedge connecting a subset of nodes. The local correlation considers neighborhood nodes that are spatially adjacent and similar in the same CNN feature maps of magnetic resonance (MR) image; and the global correlation is searched from a batch of CNN feature maps of MR images in feature space. The influence of these two correlations on semantic segmentation is complementary. We validated our LGIA module on various CNN segmentation models with the cardiac MR images dataset. Experimental results demonstrate that our approach outperformed several baseline models.     

RGB-D语义分割：深度信息的选择使用

目的 在室内场景语义分割任务中,深度信息会在一定程度上提高分割精度。但是如何更有效地利用深度信息仍是一个开放性问题。当前方法大都引入全部深度信息,然而将全部深度信息和视觉特征组合在一起可能对模型产生干扰,原因是仅依靠视觉特征网络模型就能区分的不同物体,在引入深度信息后可能产生错误判断。此外,卷积核固有的几何结构限制了卷积神经网络的建模能力,可变形卷积(deformable convolution, DC)在一定程度上缓解了这个问题。但是可变形卷积中产生位置偏移的视觉特征空间深度信息相对不足,限制了进一步发展。基于上述问题,本文提出一种深度信息引导的特征提取(depth guided feature extraction, DFE)模块。方法 深度信息引导的特征提取模块包括深度信息引导的特征选择模块(depth guided feature selection, DFS)和深度信息嵌入的可变形卷积模块(depth embedded deformable convolution, DDC)。DFS可以筛选出关键的深度信息,自适应地调整深度信息引入视觉特征的比例,在网络模型需要时将深度信息嵌...     

结合上下文编码与特征融合的SAR图像分割

目的 图像分割的中心任务是寻找更强大的特征表示,而合成孔径雷达(synthetic aperture radar, SAR)图像中斑点噪声阻碍特征提取。为加强对SAR图像特征的提取以及对特征充分利用,提出一种改进的全卷积分割网络。方法 该网络遵循编码器—解码器结构,主要包括上下文编码模块和特征融合模块两部分。上下文编码模块(contextual encoder module, CEM)通过捕获局部上下文和通道上下文信息增强对图像的特征提取;特征融合模块(feature fusion module, FFM)提取高层特征中的全局上下文信息,将其嵌入低层特征,然后将增强的低层特征并入解码网络,提升特征图分辨率恢复的准确性。结果 在两幅真实SAR图像上,采用5种基于全卷积神经网络的分割算法作为对比,并对CEM与CEM-FFM分别进行实验。结果显示,该网络分割结果的总体精度(overall accuracy, OA)、平均精度(average accuracy, AA)与Kappa系数比5种先进算法均有显著提升。其中,网络在OA上表现最好,CEM在两幅SAR图像上OA分别为91.082%和90...     

基于对抗式扩张卷积的多尺度人群密度估计

目的 人群密度估计任务是通过对人群特征的提取和分析,估算出密度分布情况和人群计数结果。现有技术运用的CNN网络中的下采样操作会丢失部分人群信息,且平均融合方式会使多尺度效应平均化,该策略并不一定能得到准确的估计结果。为了解决上述问题,提出一种新的基于对抗式扩张卷积的多尺度人群密度估计模型。方法 利用扩张卷积在不损失分辨率的情况下对输入图像进行特征提取,且不同的扩张系数可以聚集多尺度上下文信息。最后通过对抗式损失函数将网络中提取的不同尺度的特征信息以合作式的方式融合,得到准确的密度估计结果。结果 在4个主要的人群计数数据集上进行对比实验。在测试阶段,将测试图像输入训练好的生成器网络,输出预测密度图;将密度图积分求和得到总人数,并以平均绝对误差（MAE）和均方误差（MSE）作为评价指标进行结果对比。其中,在ShanghaiTech数据集上Part_A的MAE和MSE分别降至60.5和109.7,Part_B的MAE和MSE分别降至10.2和15.3,提升效果明显。结论 本文提出了一种新的基于对抗式扩张卷积的多尺度人群密度估计模型。实验结果表明,在人群分布差异较大的场景中构建的算法模型有较好的自适应性,能根据不同的场景提取特征估算密度分布,并对人群进行准确计数。     

Automatic defect detection of texture surface with an efficient texture removal network

Automated defect inspection of texture surface is still a challenging task in the industrial automation field due to the tremendous changes in the appearance of various surface textures. We present a simple but powerful image transformation network to remove textures and highlight defects at full resolution. The simple full convolution network consists only of 3 × 3 regular convolution and several dilated convolution blocks, which makes it compact and able to capture multi-scale features effectively. To further improve the ability of the network to suppress texture and highlight defects, a polynomial loss function combining perceptual loss, structural similarity loss and image gradient loss is proposed. In addition, a semi-automatic annotation method mainly composed of wavelet transform and relative total variation is designed to generate a data set of image pairs containing the original texture image and the desired texture removal image. We conducted experiments on a milled metal surface defect dataset and an open data set containing various textured backgrounds to evaluate the performance of our method. Compared with other convolutional neural network approaches, the results demonstrate the superiority of the proposed method. The method has been applied to the surface defect online detection system of an aluminum ingot milling production line, which effectively improves the surface inspection efficiency and product quality.     

CNN parameter design based on fault signal analysis and its application in bearing fault diagnosis

As a representative deep learning network, Convolutional Neural Network (CNN) has been extensively used in bearing fault diagnosis and many good results have been reported. In Prognostics and Health Management (PHM) field, the CNN’s input size is usually designed as a 1D vector or 2D square matrix, and the convolution kernel size is also defined as a square shape like 3 × 3 and 5 × 5, which are directly adopted from the image recognition. Though satisfying results can be obtained, CNN with such parameter specifications is not optimal and efficient. To this end, this paper elaborated the physical characteristics of bearing acceleration signals to guide the CNN design. First, the fault period under different fault types and shaft rotation frequency were used to determine the size of CNN’s input. Next, an exponential function was involved in fitting the envelope of decaying acceleration signal during each fault period, and signal length within different decaying ratios was used to define the CNN’s kernel size. Finally, the designed CNN was validated with the Case Western Reserve University bearing dataset and Paderborn University bearing dataset. Results confirm that the physics-guided CNN (PGCNN) with rectangular input shape and rectangular convolution kernel works better than the baseline CNN with higher accuracy and smaller uncertainty. The feasibility of designing CNN parameters with physics-guided rules derived from bearing fault signal analysis has also been verified.     

双核压缩激活神经网络艺术图像分类

目的 为了充分提取版画、中国画、油画、水彩画和水粉画等艺术图像的整体风格和局部细节特征,实现计算机自动分类检索艺术图像的需求,提出通过双核压缩激活模块（double kernel squeeze-and-excitation,DKSE）和深度可分离卷积搭建卷积神经网络对艺术图像进行分类。方法 根据SKNet（selective kernel networks）自适应调节感受野提取图像整体与细节特征的结构特点和SENet（squeeze-and-excitation networks）增强通道特征的特点构建DKSE模块,利用DKSE模块分支上的卷积核提取输入图像的整体特征与局部细节特征;将分支上的特征图进行特征融合,并对融合后的特征图进行特征压缩和激活处理;将处理后的特征加权映射到不同分支的特征图上并进行特征融合;通过DKSE模块与深度可分离卷积搭建卷积神经网络对艺术图像进行分类。结果 使用本文网络模型对有无数据增强（5类艺术图像数据增强后共25 634幅）处理的数据分类,数据增强后的分类准确率比未增强处理的准确率高9.21%。将本文方法与其他网络模型和传统分类方法相比,本文方法的分类准确率达到86.55%,比传统分类方法高26.35%。当DKSE模块分支上的卷积核为1×1和5×5,且放在本文网络模型第3个深度可分离卷积后,分类准确率达到87.58%。结论 DKSE模块可以有效提高模型分类性能,充分提取艺术图像的整体与局部细节特征,比传统网络模型具有更好的分类准确率。     

Affine Invariant Texture Segmentation and Shape from Texture by Variational Methods

We address the problem of texture segmentation by using a novel affine invariant model. The introduction of affine invariance as a requirement for texture analysis goes beyond what is known of the human performance and also beyond the psychophysical theories. We propose to compute texture features using affine invariant intrinsic neighborhoods and affine invariant intrinsic orientation matrices. We discuss several possibilities for the definition of the channels and give comparative experimental results where an affine invariant Mumford-Shah type energy functional is used to compute the multichannel affine invariant segmentation. We prove that the method is able to retrieve faithfully the texture regions and to recover the shape from texture information in images where several textures are present. The numerical algorithm is multiscale.     

Generating High-quality Superpixels in Textured Images

Superpixel segmentation is important for promoting various image processing tasks. However, existing methods still have difficulties in generating high-quality superpixels in textured images, because they cannot separate textures from structures well. Though texture filtering can be adopted for smoothing textures before superpixel segmentation, the filtering would also smooth the object boundaries, and thus weaken the quality of generated superpixels. In this paper, we propose to use the adaptive scale box smoothing instead of the texture filtering to obtain more high-quality texture and boundary information. Based on this, we design a novel distance metric to measure the distance between different pixels, which considers boundary, color and Euclidean distance simultaneously. As a result, our method can achieve high-quality superpixel segmentation in textured images without texture filtering. The experimental results demonstrate the superiority of our method over existing methods, even the learning-based methods. Benefited from using boundaries to guide superpixel segmentation, our method can also suppress noise to generate high-quality superpixels in non-textured images.     

融合逆密度函数与关系形状卷积神经网络的点云分析

目的 3D点云与以规则的密集网格表示的图像不同,不仅不规则且无序,而且由于输入输出大小和顺序差异,具有密度不均匀以及形状和缩放比例存在差异的特性。为此,提出一种对3D点云进行卷积的方法,将关系形状卷积神经网络（relation-shape convolution neural network,RSCNN）与逆密度函数相结合,并在卷积网络中增添反卷积层,实现了点云更精确的分类分割效果。方法 在关系形状卷积神经网络中,将卷积核视为由权重函数和逆密度函数组成的3D点局部坐标的非线性函数。对给定的点,权重函数通过多层感知器网络学习,逆密度函数通过核密度估计（kernel density estimation,KDE）学习,逆密度函数的引入对点云采样率不均匀的情况进行弥补。在点云分割任务中,引入由插值和关系形状卷积层两部分组成的反卷积层,将特征从子采样点云传播回原始分辨率。结果 在ModelNet40、ShapeNet、ScanNet数据集上进行分类、部分分割和语义场景分割实验,验证模型的分类分割性能。在分类实验中,与PointNet++相比,整体精度提升3.1%,在PointNet++将法线也作为输入的情况下,精度依然提升了1.9%;在部分分割实验中,类平均交并比（mean intersection over union,mIoU）比PointNet++在法线作为输入情况下高6.0%,实例mIoU比PointNet++高1.4%;在语义场景分割实验中,mIoU比PointNet++高13.7%。在ScanNet数据集上进行不同步长有无逆密度函数的对比实验,实验证明逆密度函数将分割精度提升0.8%左右,有效提升了模型性能。结论 融合逆密度函数的关系形状卷积神经网络可以有效获取点云数据中的局部和全局特征,并对点云采样不均匀的情况实现一定程度的补偿,实现更优的分类和分割效果。     

结合深度学习和支持向量机的海马子区图像分割

目的 由于海马子区体积很小且结构复杂,传统的分割方法无法达到理想的分割效果,为此提出一种基于卷积神经网络和支持向量机的海马子区分割方法。方法 该方法构建一种新模型,将卷积神经网络和支持向量机结合起来,使用支持向量机分类器替换卷积神经网络的输出层,通过训练深层网络自动提取图像块特征,利用所提取的图像特征训练支持向量机实现图像的像素级分类。结果 实验选取美国旧金山CIND中心的32位实验者的脑部磁共振图像（MRI）进行海马子区分割测试,在定性和定量方面分别对比了本文方法与支持向量机（SVM）、卷积神经网络（CNN）和基于稀疏表示与字典学习方法的分割结果。所提方法对海马子区CA1、CA2、DG、CA3、Head、Tail、SUB、ERC和PHG的分割准确率分别为0.969、0.733、0.967、0.837、0.981、0.920、0.972、0.968和0.976。本文方法优于现有的基于稀疏表示与字典学习、支持向量机和卷积神经网络的方法,各海马子区分割准确率均有较大提升,对较大子区如Head,准确率较现有最优方法提升10.2%,对较小子区如CA2、CA3,准确率分别有36.2%和52.7%的大幅提升。结论 本文方法有效提升了海马子区的分割准确率,可用于大脑核磁共振图像中海马及其子区的准确分割,为诸多神经退行性疾病的临床诊断与治疗提供依据。     

Detection of Bone Tumours in Radiographic Images using Neural Networks

We develop an approach for segmenting radiographic images of focal bone lesions possibly caused by bone tumour. A neural network is used to classify individual pixels by a convolution operation based on a feature vector. We design eight features that characterise the local texture in the neighbourhood of a pixel. Four of the features are based on co-occurrence matrices computed from the neighbourhood. The true class label of the pixels in the radiographs are obtained from annotations made by an experienced radiologist. Neural networks and self-organising feature maps are trained to perform the segmentation task. The experiments confirm the feasibility of using a feature-based neural network for finding pathologic bone changes in radiographic images. An analysis of the eight features indicates that the presence of edges and transitions, the complexity of the texture, as well as the amount of high frequencies in the texture, are the main features discriminating (soft) tissue from pathologic bone, the two classes most likely to be confused. Receiveed: 4 June 1998?,Received in revised form: 15 September 1999?Accepted: 11 December 1998     

Patch2Vec: Globally Consistent Image Patch Representation

Many image editing applications rely on the analysis of image patches. In this paper, we present a method to analyze patches by embedding them to a vector space, in which the Euclidean distance reflects patch similarity. Inspired by Word2Vec, we term our approach Patch2Vec. However, there is a significant difference between words and patches. Words have a fairly small and well defined dictionary. Image patches, on the other hand, have no such dictionary and the number of different patch types is not well defined. The problem is aggravated by the fact that each patch might contain several objects and textures. Moreover, Patch2Vec should be universal because it must be able to map never‐seen‐before texture to the vector space. The mapping is learned by analyzing the distribution of all natural patches. We use Convolutional Neural Networks (CNN) to learn Patch2Vec. In particular, we train a CNN on labeled images with a triplet‐loss objective function. The trained network encodes a given patch to a 128D vector. Patch2Vec is evaluated visually, qualitatively, and quantitatively. We then use several variants of an interactive single‐click image segmentation algorithm to demonstrate the power of our method.     

多尺度密集残差网络的单幅图像超分辨率重建

目的 近几年应用在单幅图像超分辨率重建上的深度学习算法都是使用单种尺度的卷积核提取低分辨率图像的特征信息,这样很容易造成细节信息的遗漏。另外,为了获得更好的图像超分辨率重建效果,网络模型也不断被加深,伴随而来的梯度消失问题会使得训练时间延长,难度加大。针对当前存在的超分辨率重建中的问题,本文结合GoogleNet思想、残差网络思想和密集型卷积网络思想,提出一种多尺度密集残差网络模型。方法 本文使用3种不同尺度卷积核对输入的低分辨率图像进行卷积处理,采集不同卷积核下的底层特征,这样可以较多地提取低分辨率图像中的细节信息,有利于图像恢复。再将采集的特征信息输入残差块中,每个残差块都包含了多个由卷积层和激活层构成的特征提取单元。另外,每个特征提取单元的输出都会通过短路径连接到下一个特征提取单元。短路径连接可以有效地缓解梯度消失现象,加强特征传播,促进特征再利用。接下来,融合3种卷积核提取的特征信息,经过降维处理后与3×3像素的卷积核提取的特征信息相加形成全局残差学习。最后经过重建层,得到清晰的高分辨率图像。整个训练过程中,一幅输入的低分辨率图像对应着一幅高分辨率图像标签,这种端到端的学习方法使得训练更加迅速。结果 本文使用两个客观评价标准PSNR（peak signal-to-noise ratio）和SSIM（structural similarity index）对实验的效果图进行测试,并与其他主流的方法进行对比。最终的结果显示,本文算法在Set5等多个测试数据集中的表现相比于插值法和SRCNN算法,在放大3倍时效果提升约3.4 dB和1.1 dB,在放大4倍时提升约3.5 dB和1.4 dB。结论 实验数据以及效果图证明本文算法能够较好地恢复低分辨率图像的边缘和纹理信息。     

基于级联卷积神经网络的作物病害叶片分割

针对传统卷积神经网络在作物病害叶片图像中分割精度低的问题,提出一种基于级联卷积神经网络（Cascade Convolutional Neural Network,CCNN）的作物病害叶片图像分割方法。该网络由区域病斑检测网络和区域病斑分割网络组成。基于传统VGG16模型构建区域病斑检测网络（Regional Detection Network,RD-net）,利用全局池化层代替全连接层,由此减少模型参数,实现叶片病斑区域精确定位。基于Encoder-Decoder模型结构建立区域分割网络（Regional Segmentation Network,RS-net）,并利用多尺度卷积核提高原始卷积核的局部感受野,对病斑区域精确分割。在不同环境下的病害叶片图像上进行分割实验,分割精度为87.04%、召回率为78.31%、综合评价指标值为88.22%、单幅图像分割速度为0.23?s。实验结果表明该方法能够满足不同环境下的作物病害叶片图像分割需求,可为进一步的作物病害识别方法研究提供参考。     

Polarimetric synthetic aperture radar image segmentation by convolutional neural network using graphical processing units

Image segmentation is an important application of polarimetric synthetic aperture radar. This study aimed to create an 11-layer deep convolutional neural network for this task. The Pauli decomposition formed the RGB image and was used as the input. We created an 11-layer convolutional neural network (CNN). L-band data over the San Francisco bay area and C-band data over Flevoland area were employed as the dataset. For the San Francisco bay PSAR image, our method achieved an overall accuracy of 97.32%, which was at least 2% superior to four state-of-the-art approaches. We provided the confusion matrix over test area, and the kernel visualization. We compared the max pooling and average pooling. We validated by experiment that four convolution layers perform the best. Besides, our method gave better results than AlexNet. The GPU yields a 173× acceleration on the training samples, and a 181× acceleration on the test samples, compared to standard CPU. For the Flevoland PSAR image, our 11-layer CNN also gives better overall accuracy than five state-of-the-art approaches. The convolutional neural network is better than traditional classifiers and is effective in remote sensing image segmentation.