基于流形学习的整体正交稀疏保留鉴别分析

稀疏保留投影是一种有效的特征提取方法,但是其主要关注样本间的全局稀疏重构关系,并且得到的投影变换通常不是正交的。在实际应用中,图像数据往往处于高维空间中的一种低维流形中,正交性一直被认为有利于提高鉴别能力。文中以有监督学习的方式在稀疏保留投影中引入了流形结构保留,并使得投影空间正交,从而提出了一种新的特征提取方法,即基于流形学习的整体正交稀疏保留鉴别分析（MLHOSDA）。在人脸和掌纹图像数据库的实验结果表明此方法具有较好的识别效果。     

Image decomposing for inpainting using compressed sensing in DCT domain

Inpainting images with occlusion or corruption is a challenging task. Most existing algorithms are pixel based, which construct a statistical model from image features. However, in these algorithms, the frequency component is not sufficiently addressed. In this paper, we propose a novel algorithm that utilizes compressed sensing （CS） in frequency domain to reconstruct corrupted images. In order to reconstruct image, we first decompose the image into two functions with different basic characteristics - structure component and textual component. We seek a sparse representation for the functions and use the DCT coefficients of this representation to generate an over-complete dictionary. Experimental results on real world datasets demonstrate the efficacy of our method in image inpainting. We compare our method with three state-of-the-art inpalnting algorithms and demonstrate its advantages in terms of both quantitative and qualitative aspects.     

由粗到精的虹膜图像离焦模糊评价方法

离焦模糊评价在虹膜识别系统中尤为重要。传统的方法是通过频谱分析测量虹膜图像的离焦模糊程度,这类方法容易受到光照变化以及睫毛和眼皮等噪声区域的影响。提出了一种由粗到精的虹膜图像离焦模糊评价方法。第一步,通过频谱分析去除严重模糊的虹膜图像,进行虹膜图像离焦模糊粗分类。第二步,通过方向金字塔分解,提取虹膜图像的质量特征。在人工合成的离焦模糊虹膜图像数据库中,利用径向基神经网络建立起质量特征与质量等级间的对应关系。通过建立起的模型进行实际的虹膜图像离焦模糊等级预测,以及虹膜图像离焦模糊精分类。在Clarkson数据库上的实验结果证明了该方法不仅可以准确区分清晰图像和离焦模糊图像,而且相比于传统的虹膜图像离焦评价方法更接近于人的视觉感知。     

Local adaptive segmentation algorithm for 3-D medical image based on robust feature statistics

Medical image segmentation is of pivotal importance in computer-aided clinical diagnosis. Many factors, including noises, bias field effect, local volume effect, as well as tissue movement may affect the med- ical image, thus causing blurring or uneven characteristics when forming a picture. Such quality defects will inevitably impair the gray-scale difference between adjacent tissues and lead to insufficient segmentation or even leakage during tissue or organ segmentation. In the present investigation, a local adaptive segmentation algorithm for 3-D medical image based on robust feature statistics （LARFS） was proposed. By combining segmentation algorithm principles for traditional region growing （RG） and robust feature statistics （RFS）, the location and neighborhood image information of input seed point can be comprehensively analyzed by LARFS. Results show that, for different segmentation objects, under controlling the input parameter of growing factor within certain range, LARFS segmentation algorithm can adapt well to the regional geometric shape. And be- cause the robust feature statistics is applied in the contour evolution process, LARFS algorithm is not sensitive to noises and not easily influenced by image contrast and object topology. Hence, the leakage and excessive segmentation effects are ameliorated with a smooth edge, and the accuracy can be controlled within the effective error range.     

Adaptive variational models for image decomposition

Image decomposition refers to the splitting of an image into two or more components.In this paper,a clean image is separated into two parts:one is the cartoon component,consisting only of geometric structure,and the other is the oscillatory component,consisting of texture.Three parts for noisy image are considered:cartoon,texture,and noise.To better decompose an image,we propose two new variational models.In our models,two adaptive regularization terms are introduced.The two regularization terms are determined by an adaptive function which can discriminate the cartoon and texture of an image automatically.Experimental results illustrate the efectiveness of the proposed models for image decomposition.     

Variational PCA fusion for resolution Pan-sharpening very high imagery

The Pan-sharpening approach based on principle component analysis （PCA） is affected by severe spectral distortion. To address this problem, a new pan-sharpening model based on PCA and variational technique is proposed to construct the substitute image of the first principal component （PC1）. The energy functional consists of three terms. The first term injects PC1 with the geometric structure of the panchromatic （Pan） image. The second term preserves the spectral pattern of the multi-spectral image in the merged result. And the third term guarantees the smoothness of the functional optimization solution. The fusion result is given by the minimum of the energy functional, which is computed with the gradient descend flow. The experiments on QuickBird and IKONOS datasets validate the effectiveness of the proposed model. Compared with the state- of-the-art pan-sharpening approaches, this model exhibits a better trade-off between improving spatial quality and preserving spectral signature of the MS image.     

Re-GAN：残差生成式对抗网络算法

目的 生成式对抗网络（generative adversarial network,GAN）是一种无监督生成模型,通过生成模型和判别模型的博弈学习生成图像。GAN的生成模型是逐级直接生成图像,下级网络无法得知上级网络学习的特征,以至于生成的图像多样性不够丰富。另外,随着网络层数的增加,参数变多,反向传播变得困难,出现训练不稳定和梯度消失等问题。针对上述问题,基于残差网络（residual network,ResNet）和组标准化（group normalization,GN）,提出了一种残差生成式对抗网络（residual generative adversarial networks,Re-GAN）。方法 Re-GAN在生成模型中构建深度残差网络模块,通过跳连接的方式融合上级网络学习的特征,增强生成图像的多样性和质量,改善反向传播过程,增强生成式对抗网络的训练稳定性,缓解梯度消失。随后采用组标准化（GN）来适应不同批次的学习,使训练过程更加稳定。结果 在Cifar10、CelebA和LSUN数据集上对算法的性能进行测试。Re-GAN的IS （inception score）均值在批次为64时,比DCGAN （deep convolutional GAN）和WGAN （Wasserstein-GAN）分别提高了5%和30%,在批次为4时,比DCGAN和WGAN分别提高了0.2%和13%,表明无论批次大小,Re-GAN生成图像具有很好的多样性。Re-GAN的FID （Fréchet inception distance）在批次为64时比DCGAN和WGAN分别降低了18%和11%,在批次为4时比DCGAN和WGAN分别降低了4%和10%,表明Re-GAN生成图像的质量更好。同时,Re-GAN缓解了训练过程中出现的训练不稳定和梯度消失等问题。结论 实验结果表明,在图像生成方面,Re-GAN的生成图像质量高、多样性强;在网络训练方面,Re-GAN在不同批次下的训练具有更好的兼容性,使训练过程更加稳定,梯度消失得到缓解。     

基于FCN-LSTM的工业烟尘图像分割

工业生产中常根据林格曼烟气黑度判断工业烟尘的污染等级,一种有效的方式是应用计算机视觉系统对工业烟尘进行监测, 其中对烟尘目标进行准确分割是该系统的关键技术。因为工业烟尘具有形状不固定、和云相似度高等特点,现有算法在复杂场景下对烟尘进行分割时容易受到干扰,分割准确度有待提高。针对这一问题,提出一种基于FCN-LSTM的工业烟尘图像分割方法,在全卷积网络对图像空间特征提取的基础上,使用长短时记忆网络提取图像序列的时间信息,通过烟尘的动态特征对运动的烟尘和背景进行区分,增强复杂场景下的抗干扰能力。实验表明,本文模型相比于全卷积网络,在复杂场景下的抗干扰能力有显著提升,能够有效克服来自云的干扰,对全卷积网络分割结果中易出现干扰点的问题也有改善,IoU指标最高有8.04%的提升。     

应用型本科院校“数字图像处理”教学研究与改进

"数字图像处理"是电子信息类专业的基础课程具有很强理论性和工程应用性。根据培养高素质应用型人才的培养目标以及课程思政的要求,分析当前教学模式中存在的问题和不足,提出旨在提高教学质量和人才培养质量的方法和措施。教学改革措施增强了学生对该课程的学习兴趣和主动性,培养了学生的爱国情怀、创新思维,提升了学生的实际工程能力,且提高了教学质量。