EDU-GAN:Edge Enhancement Generative Adversarial Networks with Dual-Domain Discriminators for Inscription Images Denoising

Recovering high-quality inscription images from unknown and complex inscription noisy images is a challenging research issue.Different fromnatural images,character images pay more attention to stroke information.However,existingmodelsmainly consider pixel-level informationwhile ignoring structural information of the character,such as its edge and glyph,resulting in reconstructed images with mottled local structure and character damage.To solve these problems,we propose a novel generative adversarial network(GAN)framework based on an edge-guided generator and a discriminator constructed by a dual-domain U-Net framework,i.e.,EDU-GAN.Unlike existing frameworks,the generator introduces the edge extractionmodule,guiding it into the denoising process through the attention mechanism,which maintains the edge detail of the restored inscription image.Moreover,a dual-domain U-Net-based discriminator is proposed to learn the global and local discrepancy between the denoised and the label images in both image and morphological domains,which is helpful to blind denoising tasks.The proposed dual-domain discriminator and generator for adversarial training can reduce local artifacts and keep the denoised character structure intact.Due to the lack of a real-inscription image,we built the real-inscription dataset to provide an effective benchmark for studying inscription image denoising.The experimental results show the superiority of our method both in the synthetic and real-inscription datasets.

I-DCGAN and TOPSIS-IFP:A simulation generation model for radiographic flaw detection images in light alloy castings and an algorithm for quality evaluation of generated images

The intelligent detection technology driven by X-ray images and deep learning represents the forefront of advanced techniques and development trends in flaw detection and automated evaluation of light alloy castings.However,the efficacy of deep learning models hinges upon a substantial abundance of flaw samples.The existing research on X-ray image augmentation for flaw detection suffers from shortcomings such as poor diversity of flaw samples and low reliability of quality evaluation.To this end,a novel approach was put forward,which involves the creation of the Interpolation-Deep Convolutional Generative Adversarial Network(I-DCGAN)for flaw detection image generation and a comprehensive evaluation algorithm named TOPSIS-IFP.I-DCGAN enables the generation of high-resolution,diverse simulated images with multiple appearances,achieving an improvement in sample diversity and quality while maintaining a relatively lower computational complexity.TOPSIS-IFP facilitates multi-dimensional quality evaluation,including aspects such as diversity,authenticity,image distribution difference,and image distortion degree.The results indicate that the X-ray radiographic images of magnesium and aluminum alloy castings achieve optimal performance when trained up to the 800th and 600th epochs,respectively.The TOPSIS-IFP value reaches 78.7%and 73.8%similarity to the ideal solution,respectively.Compared to single index evaluation,the TOPSIS-IFP algorithm achieves higher-quality simulated images at the optimal training epoch.This approach successfully mitigates the issue of unreliable quality associated with single index evaluation.The image generation and comprehensive quality evaluation method developed in this paper provides a novel approach for image augmentation in flaw recognition,holding significant importance for enhancing the robustness of subsequent flaw recognition networks.

Research on the Pedestrian Re-Identification Method Based on Local Features and Gait Energy Images

The appearance of pedestrians can vary greatly from image to image,and different pedestrians may look similar in a given image.Such similarities and variabilities in the appearance and clothing of individuals make the task of pedestrian re-identification very challenging.Here,a pedestrian re-identification method based on the fusion of local features and gait energy image(GEI)features is proposed.In this method,the human body is divided into four regions according to joint points.The color and texture of each region of the human body are extracted as local features,and GEI features of the pedestrian gait are also obtained.These features are then fused with the local and GEI features of the person.Independent distance measure learning using the cross-view quadratic discriminant analysis(XQDA)method is used to obtain the similarity of the metric function of the image pairs,and the final similarity is acquired by weight matching.Evaluation of experimental results by cumulative matching characteristic(CMC)curves reveals that,after fusion of local and GEI features,the pedestrian re-identification effect is improved compared with existing methods and is notably better than the recognition rate of pedestrian re-identification with a single feature.

Discrimination of Acacia seeds at species and subspecies levels using an image analyzer

Seeds of Acacia species and subspecies were characterized using an image analyzer and discriminated for the purpose of identification of species, using their seeds. The species considered in the study were Acacia nilotica subsp. indica, A. nilotica subsp. cupressiformis, A. nilotica subsp. tomentosa, A. tortilis subsp. raddiana, A. tortilis subsp. spirocarpa, A. raddiana, A. senegal, A. auriculiformis, A. farnesiana, A. leucophloea, A. mearnsii, A. melanoxylon, A. planifrons and A. mangium. Eight samples each consisting of 25 seeds per species were studied using the image analyzer for physical characteristics of seeds, such as 2D surface area, length, width, perimeter, roundness, aspect ratio and fullness ratio. Discriminant analysis showed that acacias can be discriminated at species and subspecies levels, with 96% accuracy. Exceptions were A. nilotica subsp. tomentosa(75.0%), A. tortilis subsp. spirocarpa(75.0%) and A. raddiana(87.5%) which had relatively low discrimination accuracy. However, discriminant analysis within selected species showed complete recognition of these species except for A. tortilis subsp. spirocarpa, that had still a large overlap with A. leucophloea. The study also revealed that both seed size and shape characteristics were responsible for species discrimination. It can be concluded that rapid analysis of seed size and shape characteristics using image analysis techniques can be used as primary and secondary keys for identification of acacias.

Image Processing System for Air Classification Using Linear Discriminant Analysis

An air classifier is used in the recycling process of covered electric wire in the recycling factories, in which the covered electric wires are crushed, sieved, and classified by the air classifier, which generates wastes. In these factories, operators manually adjust the air flow rate while checking the wastes discharged from the separator outlet. However, the adjustments are basically done by trial and error, and it is difficult to do them appropriately. In this study, we tried to develop the image processing system that calculates the ratio of copper (Cu) product and polyvinyl chloride (PVC) in the wastes as a substitute for the operator’s eyes. Six colors of PVC (white, gray, green, blue, black, and red) were used in the present work. An image consists of foreground and background. An image’s regions of interest are objects (Cu particles) in its foreground. However, the particles having a color similar to the background color are buried in the background. Using the difference of two color backgrounds, we separated particles and background without dependent of background. The Otsu’ thresholding was employed to choose the threshold to maximize the degree of separation of the particles and background. The ratio of Cu to PVC pixels from mixed image was calculated by linear discriminant analysis. The error of PVC pixels resulted in zero, whereas the error of Cu pixels arose to 4.19%. Comparing the numbers of Cu and PVC pixels within the contour, the minority of the object were corrected to the majority of the object. The error of Cu pixels discriminated as PVC incorrectly became zero percent through this correction.

A New Shadow Removal Method for Color Images

Shadow and variable illumination considerably influence the results of image understanding such as image segmentation, object tracking, and object recognition. The intrinsic image decomposition is to separate the reflectance and the illumination image from an observed image. The intrinsic image decomposition is very useful to remove shadows and then improve the performance of image understanding. In this paper, we present a new shadow removal method based on intrinsic image decomposition on a single color image using the Fisher Linear Discriminant (FLD). Under the assumptions-Lambertian surfaces, approximately Planckian lighting, and narrowband camera sensors, there exist an invariant image, which is 1-dimensional greyscale and independent of illuminant color and intensity. The Fisher Linear Discriminant is applied to create the invariant image. And further the shadows can be removed through the difference between invariant image and original color image. The experimental results on real data show good performance of this algorithm.

Multispectral Imaging in Combination with Multivariate Analysis Discriminates Selenite Induced Cataractous Lenses from Healthy Lenses of Sprague-Dawley Rats

Cataracts are the leading cause of blindness worldwide. Current methods for discriminating cataractous lenses from healthy lenses of Sprague-Dawley rats during preclinical studies are based on either histopathological or clinical assessments which are weakened by subjectivity. In this work, both cataractous and healthy lens tissues of Sprague-Dawley rats were studied using multispectral imaging technique in combination with multivariate analysis. Multispectral images were captured in transmission, reflection and scattering modes. In all, five spectral bands were found to be markers for discriminating cataractous lenses from healthy lenses;470 nm and 625 nm discriminated in reflection mode whereas 435 nm, 590 nm and 700 nm discriminated in transmission mode. With Fisher’s Linear discriminant analysis, the midpoints for classifying cataractous from healthy lenses were found to be 14.718 × 10−14 and 3.2374 × 10−14 for the two spectra bands in the reflection mode and the three spectral bands in the transmission mode respectively. Images in scattering mode did not show significant discrimination. These spectral bands in reflection and transmission modes may offer potential diagnostic markers for discriminating cataractous lenses from healthy lenses thereby promising multispectral imaging applications for characterizing cataractous and healthy lenses.

Perceptual Image Outpainting Assisted by Low-Level Feature Fusion and Multi-Patch Discriminator

Recently,deep learning-based image outpainting has made greatly notable improvements in computer vision field.However,due to the lack of fully extracting image information,the existing methods often generate unnatural and blurry outpainting results in most cases.To solve this issue,we propose a perceptual image outpainting method,which effectively takes the advantage of low-level feature fusion and multi-patch discriminator.Specifically,we first fuse the texture information in the low-level feature map of encoder,and simultaneously incorporate these aggregated features reusability with semantic(or structural)information of deep feature map such that we could utilizemore sophisticated texture information to generate more authentic outpainting images.Then we also introduce a multi-patch discriminator to enhance the generated texture,which effectively judges the generated image from the different level features and concurrently impels our network to produce more natural and clearer outpainting results.Moreover,we further introduce perceptual loss and style loss to effectively improve the texture and style of outpainting images.Compared with the existing methods,our method could produce finer outpainting results.Experimental results on Places2 and Paris StreetView datasets illustrated the effectiveness of our method for image outpainting.

Differentiation of Wheat Diseases and Pests Based on Hyperspectral Imaging Technology with a Few Specific Bands

Hyperspectral imaging technique is known as a promising non-destructive way for detecting plants diseases and pests.In most previous studies,the utilization of the whole spectrum or a large number of bands as well as the complexity of model structure severely hampers the application of the technique in practice.If a detection system can be established with a few bands and a relatively simple logic,it would be of great significance for application.This study established a method for identifying and discriminating three commonly occurring diseases and pests of wheat,i.e.,powdery mildew,yellow rust and aphid with a few specific bands.Through a comprehensive spectral analysis,only three bands at 570,680 and 750 nm were selected.A novel vegetation index namely Ratio Triangular Vegetation Index(RTVI)was developed for detecting anomalous areas on leaves.Then,the Support Vector Machine(SVM)method was applied to construct the discrimination model based on the spectral ratio analysis.The validating results suggested that the proposed method with only three spectral bands achieved a promising accuracy with the Overall Accuracy(OA)of 83%.With three bands from the hyperspectral imaging data,the three wheat diseases and pests were successfully detected and discriminated.A stepwise strategy including background removal,damage lesions recognition and stresses discrimination was proposed.The present work can provide a basis for the design of low cost and smart instruments for disease and pest detection.

基于太赫兹成像检测技术与特征提取方法结合巴旦木饱满度检测方法研究

巴旦木是一种营养丰富的坚果,对巴旦木的品质进行检测具有重要的经济价值和实际意义。由于巴旦木具有较为坚硬的外壳,传统的检测手段较难实现内部检测,因此,采用新兴的太赫兹透射成像检测技术,开展巴旦木饱满度的检测研究。首先采集不同饱满度巴旦木的太赫兹透射图像,并且从太赫兹图像的感兴趣区域分别提取无样品区域、空壳区域和满仁区域的太赫兹光谱信息;为了提高模型的精度,减少计算量,采用竞争性自适应重加权算法(CARS)、无信息变量消除(UVE)、连续投影算法(SPA)、蒙特卡罗无信息变量消除法(MCUVE)和遗传算法(GA)对太赫兹光谱信息进行特征提取,建立对应的最小二乘支持向量机(LS-SVM)、随机森林(RF)和K-近邻(KNN)定性判别模型,对巴旦木的饱满和空壳区域进行检测和鉴别。此外,对太赫兹特征图像转为JPG格式,接着转化为RGB格式进行G通道提取和图像二值化分离出外壳和果仁图像,检测饱满度为太赫兹特征图像的壳仁像素点之比;对原始图像进行轮廓提取和图像二值化分离出外壳和果仁图像,实际饱满度为原始图像的壳仁像素点之比。通过计算检测饱满度和实际饱满度的误差,证明了太赫兹透射成像技术检测巴旦木饱满度的可行性。建立的KS-GA-RF模型的鉴别效果最优,准确率为98.21%;通过壳仁像素点之比分别计算出对应的检测饱满度和实际饱满度,误差为16%。研究验证了采用太赫兹图、谱相融合的方法,可以很好地实现对巴旦木内部种仁饱满度可视化检测,为巴旦木的准确分级提供了新的思路,也为太赫兹成像技术检测其他坚果饱满度提供了理论参考,具有重要的应用价值。

薄层CT对肺结节性质鉴别的诊断价值分析

目的:探讨薄层CT影像学特征在肺结节性质鉴别中的诊断价值。方法:回顾性分析100例肺结节患者的临床资料,依据病理检测结果将其分为良性组和恶性组,所有患者均接受薄层CT影像学检查。观察并比较2组患者的结节征象、分级评估结果、薄层CT影像学特征出现情况及薄层CT影像学指标;通过Logistic回归分析讨论各指标对肺结节良恶性诊断的影响,通过受试者操作特征曲线(ROC曲线)分析各指标对肺结节良恶性诊断的价值。结果:2组患者在肺结节分级系统评估结果、薄层CT影像学特征、结节相关指标方面差异均有统计学意义(P<0.05)。Logistic回归分析结果显示,肺结节分级、结节最大直径、平均CT值、最大CT值是影响恶性肺结节诊断的独立因素(P<0.05),随着分级级别升高和薄层CT影像学相关指标增加,恶性肺结节的确诊风险增加。ROC曲线分析结果表明,肺结节分级、结节最大直径、平均CT值、最大CT值在肺结节良恶性诊断中均具有明确的诊断价值,尤其是联合预测因子的诊断效能最为显著。结论:薄层CT影像学特征对肺结节性质鉴别具有重要的临床意义,综合考虑肺结节分级、结节最大直径、平均CT值、最大CT值能够更准确地判断肺结节的性质,为医生提供更全面的临床参考。

多路径生成对抗网络的红外与可见光图像融合

生成对抗网络在红外与可见光图像融合领域受到广泛关注,但单路径进行融合容易丢失浅层信息、分支路特征提取融合能力有限。提出一种基于多路径生成对抗网络的红外与可见光图像融合方法。在生成器端,利用源图像与导向滤波结果构建3条输入路径提取更多源图像特征信息,以获得细节更丰富的融合图像;然后,卷积层加入掩码注意力机制模块,提升显著信息的提取效率,引入密集连接和残差连接,在提升特征传递效率的同时可获取更多源图像重要特征信息。在鉴别器端,采用双鉴别器估计红外与可见光图像的区域分布,避免单鉴别器网络丢失对比度信息的模态失衡问题。在TNO数据集上进行了实验,实验结果表明,所提算法在5个客观评估指标上4项取得了最好结果,优于多数主流算法,在主观评估方面,所提算法保留了更多的纹理细节信息,具有更好的视觉效果。

半监督空谱局部判别分析的高光谱影像特征提取

为充分利用高光谱影像中蕴含的空谱特征,提出了一种半监督空谱局部判别分析的高光谱影像特征提取算法(S4LFDA)。鉴于高光谱数据集具有空间一致性,首先将像元进行空间重构,保存高光谱数据的近邻关系;其次引入光谱信息散度重构像元间的相似度;为了充分利用大量无标签样本提高算法性能,采用模糊C均值聚类算法对样本进行聚类分析得到伪标签;然后通过增加规范化项到局部力导引算法(FDA)的类内散度矩阵和类间散度矩阵中,以此保持无标签样本的聚类结构一致性;最后通过局部FDA算法来保持有标签样本类间散度最大化和类内散度最小化并求解最佳投影向量。S4LFDA算法既保持了数据集在光谱域的可分性,又保持了像元在空间区域内的近邻关系,合理利用有标签样本及无标签样本,提高了算法的分类性能。在Pavia University和Indian Pines数据集上进行实验,总体分类精度达到95.60%和94.38%。与其他维数约简算法相比,该算法有效提高了地物分类性能。

基于三分支对抗学习和补偿注意力的红外和可见光图像融合

针对现有深度学习图像融合方法依赖卷积提取特征,并未考虑源图像全局特征,融合结果容易产生纹理模糊、对比度低等问题,本文提出一种基于三分支对抗学习和补偿注意力的红外和可见光图像融合方法。首先,生成器网络采用密集块和补偿注意力机制构建局部-全局三分支提取特征信息。然后,利用通道特征和空间特征变化构建补偿注意力机制提取全局信息,更进一步提取红外目标和可见光细节表征。其次,设计聚焦双对抗鉴别器,以确定融合结果和源图像之间的相似分布。最后,选用公开数据集TNO和RoadScene进行实验并与其他9种具有代表性的图像融合方法进行对比,本文提出的方法不仅获得纹理细节更清晰、对比度更好的融合结果,而且客观度量指标优于其他先进方法。

基于显微特征颜色量化评价判别栀子与焦栀子

目的:建立基于显微特征颜色量化评价的栀子与焦栀子判别方法。方法:采用显微成像技术和显微特征颜色提取软件测定栀子与焦栀子显微特征颜色,利用Kruska-WallisH秩和检验、Fisher判别分析法分析栀子与焦栀子显微特征颜色差异,建立判别函数。结果:栀子与焦栀子显微特征(内果皮石细胞、内果皮纤维、种皮石细胞)的亮度值(L^(*))、红绿色值(a^(*))、黄蓝色值(b^(*))和总色度值(E_(ab)^(*))均有显著差异(P<0.01)。以种皮石细胞颜色对二者进行判别,其判别函数为y=0.678×L^(*)+0.384×a^(*)-0.576×b^(*)-9.322,y>0为栀子,y<0为焦栀子。结论:通过显微特征颜色量化评价,实现了栀子与焦栀子的有效判别,为栀子生、制饮片的判别和质量评价提供了新方法。

双鉴别器盲超分重建方法研究

图像超分变率重建方法在公共安全检测、卫星成像、医学和照片恢复等方面有着十分重要的用途。该文对基于生成对抗网络的超分辨率重建方法进行研究,提出一种基于纯合成数据训练的真实世界盲超分算法(RealESRGAN)的UNet3+双鉴别器Real-ESRGAN方法(Double Unet3+Real-ESRGAN, DU3-Real-ESRGAN)。首先,在鉴别器中引入UNet3+结构,从全尺度捕捉细粒度的细节和粗粒度的语义。其次,采用双鉴别器结构,一个鉴别器学习图像纹理细节,另一个鉴别器关注图像边缘,实现图像信息互补。在Set5, Set14, BSD100和Urban100数据集上,与多种基于生成对抗网络的超分重建方法相比,除Set5数据集外,DU3-Real-ESRGAN方法在峰值信噪比(PSNR)、结构相似性(SSIM)和无参图像考评价指标(NIQE)都优于其他方法,产生了更直观逼真的高分辨率图像。

基于Fisher判别分析可分性信息融合的马铃薯VC含量高光谱检测方法

为提高马铃薯VC含量检测结果的准确性和可靠性,提出一种基于Fisher判别分析(Fisher discriminant analysis,FDA)可分性数据融合的检测模型输入变量构建方法。首先,利用高光谱成像技术采集200个马铃薯的高光谱信息,通过对比6种预处理方法和原始数据的建模结果,确定多元散射校正为光谱数据的预处理方法;其次,采用竞争性自适应重加权采样(competitive adaptive reweighted sampling,CARS)、连续投影算法(successive projections algorithm,SPA)及CARS-SPA组合算法3种方法提取相应特征波长,通过对比分析最终确定34个有效特征波长;然后,将有效特征波长进行FDA可分性数据融合,根据融合的新变量对样本间差异性判别能力的大小进行筛选,确定构建检测模型的输入变量;最后,分别对FDA融合前后筛选的变量建立偏最小二乘模型和反向传播神经网络(back propagation neural network,BPNN)模型,并对检测结果进行对比分析。结果表明,将CARS算法提取的34个特征波长进行FDA融合,采用前3个融合变量作为构建检测模型的输入变量时,其所建BPNN模型的相关系数由0.9726提高至0.9990,均方根误差由0.7723降低至0.1727,不仅能够极大地降低数据分析维度,而且能够提高检测结果的准确性。因此,基于FDA可分性数据融合构建检测模型输入变量可以提高马铃薯VC含量检测结果的准确性。

基于改进3E-LDA的织物图像分类算法

针对训练样本数太少(训练样本数量小于数据维数)导致的模型分辨能力下降问题,提出了一种基于正则化改进3E-LDA的织物图像分类算法(I3E-LDA算法)。首先利用类加权中值代替样本均值计算类内散点矩阵,削弱离群值和噪声的影响,以此作为非参数加权特征提取法对类内散点矩阵进行正则化。然后利用目标组合的方法,通过引入平衡参数对目标函数进行正则化,来保留更具判别性的特征数据。通过不同织物图像间更具判别性的特征数据可以更好地对其区分。结合改进的零空间法解决类内散点矩阵奇异性和小样本问题,从而提高分类准确率。在阿里天池织物数据集和花色织物图像上进行训练和测试,将图像按照正常图像和非正常图形(瑕疵图像)进行区分。实验结果表明,I3E-LDA算法有效实现了织物图像分类,且对于较少的训练样本(20%~40%的样本用于训练)提升了分类精度。

级联式生成对抗网络的全景图像修复

为了解决全景图像视场宽、畸变显著等问题,提出了一种级联式生成对抗网络的全景图像修复算法。第一阶段提出了一种双判别器生成对抗网络,通过对等矩形格式的全景图像进行立方体投影转换,对立方体六面图像进行修复,引入PatchGAN作为全局判别器捕获细节信息,局部判别器网络可以保证局部修复结果与周围区域的一致性。第二阶段提出了一种失真感知生成对抗网络,通过矩形混合卷积缓解全景图像失真,判别器引入谱归一化,与第一阶段进行级联以缓解立方体图像边界不连续问题,设计联合损失函数以优化网络修复效果。实验结果表明,所提算法无论从主观视觉评价或是从客观评价指标上均取得了优秀的效果,实现全景图像的有效修复。