基于弱语义嵌入的图像相似性度量

提出了一种旨在缩小"语义鸿沟"的弱语义嵌入方法及其相似性度量模型.在该方法中,用户反馈的相关图像被当成查询图像的弱语义直接嵌入到视觉特征向量中,从而使固定的视觉特征向量成为可伸缩的包含语义和视觉特征的集合.根据心理学中集合论相似性理论,图像之间的相似性被表示成语义特征与视觉相似性之比.基于Corel图像库的实验结果表明,该方法在基于内容的图像检索中是非常有效的.     

A Semantic Supervision Method for Abstractive Summarization

In recent years, many text summarization models based on pre-training methods have achieved very good results. However, in these text summarization models, semantic deviations are easy to occur between the original input representation and the representation that passed multi-layer encoder, which may result in inconsistencies between the generated summary and the source text content. The Bidirectional Encoder Representations from Transformers (BERT) improves the performance of many tasks in Natural Language Processing (NLP). Although BERT has a strong capability to encode context, it lacks the fine-grained semantic representation. To solve these two problems, we proposed a semantic supervision method based on Capsule Network. Firstly, we extracted the fine-grained semantic representation of the input and encoded result in BERT by Capsule Network. Secondly, we used the fine-grained semantic representation of the input to supervise the fine-grained semantic representation of the encoded result. Then we evaluated our model on a popular Chinese social media dataset (LCSTS), and the result showed that our model achieved higher ROUGE scores (including R-1, R-2), and our model outperformed baseline systems. Finally, we conducted a comparative study on the stability of the model, and the experimental results showed that our model was more stable.     

Clinical evaluation of a new set of image quality criteria for mammography

The European Commission (EC) quality criteria for screen-film mammography are used as a tool to assess image quality. A new set of criteria was developed and initially tested in a previous study. In the present study, these criteria are further evaluated using screen-film mammograms that have been digitised, manipulated to simulate different image quality levels and reprinted on film. Expert radiologists have evaluated these manipulated images using both the original (EC) and the new criteria. A comparison of three different simulated dose levels reveals that the new criteria yield a larger separation of image criteria scores than the old ones. These results indicate that the new set of image quality criteria has a higher discriminative power than the old set and thus seems to be more suitable for evaluation of image quality in mammography.     

基于美学质量评价的条件生成对抗网络线稿图像彩色化方法

目的 为了更好地实现包装设计中符合用户美学感知需求的动漫图像元素,生成美感更高的彩色线稿图像,提出一种基于美学质量评价的条件生成对抗网络线稿图像彩色化方法(Conditional Image Colorization with Image Aesthetics GANs,IM-GAN).方法 采用Mish函数作为生成模型激活函数,并使用美学质量评价优化生成模型损失函数,实现线稿图像的自动彩色化任务.结果 使用图像美学质量评分作为客观评价指标,观察者打分作为主观评价指标,对算法进行评价.IM-GAN生成的彩色图像具有更高美学质量评分和主观打分.结论 文中方法能够完成线稿图像自动彩色化任务,在包装设计动漫形象应用方面具有一定的参考和使用价值.     

基于视觉显著性和感知相似性的全参考图像质量评价方法

目的 图像质量评价（IQA）旨在使用计算模型自动衡量和评价图像质量,以代替人类视觉系统的主观意见,并应用到相关实际问题中。方法 首先将参考图像与失真图像进行输入,使用视觉显著性模型计算图像局部相似度的特征映射,并在质量得分池化阶段作为加权函数,同时,针对视觉显著性图作为单一特征映射的不足,增加了梯度幅度,然后将图像进行颜色空间的转化提取颜色特征,最后分配相应的权重来计算图像相似度。结果 在4个大型数据集上的对比测试显示,在保持适度计算复杂度的同时,VSPSI相比其他有代表性的模型在预测精度上得到了一定的提升,特别是在TID2013数据集上的SROCC达到了0.9055。结论 研究结果表明,VSPSI是一个性能优良的IQA方法,在不同数据集和不同失真类型中都有良好的表现,具有较强的鲁棒性,可胜任多类失真图像的客观质量评价,同时可通过优化视觉显著性模型进一步提升VSPSI的性能。     

Inter-observer variation in masked and unmasked images for quality evaluation of clinical radiographs

PURPOSE: To investigate the influence of masking on the inter-observer variation in image quality evaluation of clinical radiographs of chest and lumbar spine. BACKGROUND: Inter-observer variation is a big problem in image quality evaluation since this variation is often much bigger than the variation in image quality between, for example, two radiographic systems. In this study, we have evaluated the effect of masking on the inter-observer variation. The idea of the masking was to force every observer to view exactly the same part of the image and to avoid the effect of the overall 'first impression' of the image. A discussion with a group of European expert radiologists before the study indicated that masking might be a good way to reduce the inter-observer variation. METHODS: Five chest and five lumbar spine radiographs were collected together with detailed information regarding exposure conditions. The radiographs were digitised with a high-performance scanner and five different manipulations were performed, simulating five different exposure conditions. The contrast, noise and spatial resolution were manipulated by this method. The images were printed onto the film and the individual masks were produced for each film, showing only the parts of the images that were necessary for the image quality evaluation. The quality of the images was evaluated on ordinary viewing boxes by a large group of experienced radiologists. The images were examined with and without the masks with a set of image criteria (if fulfilled, 1 point; and not fulfilled, 0 point), and the mean score was calculated for each simulated exposure condition. RESULTS: The results of this study indicate that-contrary to what was supposed-the inter-observer variation increased when the images were masked. In some cases, especially for chest, this increase was statistically significant. CONCLUSIONS: Based on the results of this study, image masking in studies of fulfilment of image criteria cannot be recommended.     

The spatial statistics of structural magnetic resonance images: application to post-acquisition quality assessment of brain MRI images

This report describes a new quality evaluation method for structural magnetic resonance images (MRI) of the brain. Pixels in MRI images are regarded as regionalized random variables that exhibit distinct and organized geographic patterns. We extract geo-spatial local entropy features and build three separate Gaussian distributed quality models upon them using 250 brain MRI images of different subjects. The MRI images were provided by Alzheimer's disease neuroimaging initiative (ADNI). Image quality of a test image is predicted in a three-step process. In the first step, three separate geo-spatial feature vectors are extracted. The second step standardizes each quality model using corresponding geo-spatial feature vector extracted from the test image. The third step computes image quality by transforming the standardized score to probability. The proposed method was evaluated on images without perceived distortion and images degraded by different levels of motion blur and Rician noise as well as images with different configurations of bias fields. Based on the performance evaluation, our proposed method will be suitable for use in the field of clinical research where quality evaluation is required for the brain MRI images acquired from different MRI scanners and different clinical trial sites before they are fed into automated image processing and image analysis systems.     

Rotation- and scale-invariant feature extraction by a diffractive optical inner-product transform

A diffractive optical element that is capable of extracting rotation- and scale-invariant features from input images is presented. The diffractive optical element performs an inner product between the input image and a set of basis functions. This set of basis functions is derived from that used in the Fourier-Mellin transform. Simulated results from scaled and rotated versions of two input images are presented.     

Evaluation of image quality of lumbar spine images: a comparison between FFE and VGA

PURPOSE: The aim of the present study is to compare two different methods for evaluation of the quality of clinical X-ray images. METHODS: Based on fifteen lumbar spine radiographs, two new sets of images were created. A hybrid image set was created by adding two distributions of artificial lesions to each original image. The image quality parameters spatial resolution and noise were manipulated and a total of 210 hybrid images were created. A set of 105 disease-free images was created by applying the same combinations of spatial resolution and noise to the original images. The hybrid images were evaluated with the free-response forced error experiment (FFE) and the normal images with visual grading analysis (VGA) by nine experienced radiologists. RESULTS: In the VGA study, images with low noise were preferred over images with higher noise levels. The alteration of the MTF had a limited influence on the VGA score. For the FFE study, the visibility of the lesions was independent of the sharpness and the noise level. No correlation was found between the two image quality measures. CONCLUSIONS: FFE is a precise method for evaluation of image quality, but the results are only valid for the type of lesion used in the study, whereas VGA is a more general method for clinical image quality assessment. The results of the FFE study indicate that there might be a potential to lower the dose levels in lumbar spine radiography without losing important diagnostic information.     

Efficient Facial Recognition Authentication Using Edge and Density Variant Sketch Generator

Image translation plays a significant role in realistic image synthesis, entertainment tasks such as editing and colorization, and security including personal identification. In Edge GAN, the major contribution is attribute guided vector that enables high visual quality content generation. This research study proposes automatic face image realism from freehand sketches based on Edge GAN. We propose a density variant image synthesis model, allowing the input sketch to encompass face features with minute details. The density level is projected into non-latent space, having a linear controlled function parameter. This assists the user to appropriately devise the variant densities of facial sketches and image synthesis. Composite data set of Large Scale CelebFaces Attributes (ClebA), Labelled Faces in the Wild (LFWH), Chinese University of Hong Kong (CHUK), and self-generated Asian images are used to evaluate the proposed approach. The solution is validated to have the capability for generating realistic face images through quantitative and qualitative results and human evaluation.     

基于DeepLabv3+的图像语义分割优化方法

目的为了实现良好的图像语义分割精度,同时尽可能降低网络的参数量,加快网络训练速度,提出基于DeepLabv3+的图像语义分割优化方法。方法编码器主干网络增加注意力机制模块,并采用更密集的特征池化模块有效聚合多尺度特征,同时使用深度可分离卷积降低网络计算复杂度。结果基于CamVid数据集的对比实验显示,优化后网络的MIoU分数达到了71.03%,在像素精度、平均像素精度等其他方面的评价指标上较原网络有小幅提升,并且网络参数量降低了12%。在Cityscapes的测试数据集上的MIoU分数为75.1%。结论实验结果表明,优化后的网络能够有效提取图像特征信息,提高语义分割精度,同时降低模型复杂度。文中网络使用城市道路场景数据集进行测试,可以为今后的无人驾驶技术的应用提供参考,具有一定的实际意义。     

Can the average glandular dose in routine digital mammography screening be reduced? A pilot study using revised image quality criteria

There is a need for tools that in a simple way can be used for the evaluation of image quality related to clinical requirements in mammography. The aim of this work was to adjust the present European image quality criteria to be relevant also for digital mammography images, and to use as simple and as few criteria as possible. A pilot evaluation of the new set of criteria was made with mammograms of 28 women from a General Electric Senographe 2000D full-field digital mammography system. One breast was exposed using the standard automatic exposure mode, the other using about half of that absorbed dose. Three experienced radiologists evaluated the images using visual grading analysis technique. The results indicate that the new quality criteria can be used for the evaluation of image quality related to clinical requirements in digital mammography in a simple way. The results also suggest that absorbed doses for the mammography system used may be substantially reduced.     

Sentence Similarity Measurement with Convolutional Neural Networks Using Semantic and Syntactic Features

Calculating the semantic similarity of two sentences is an extremely challenging problem. We propose a solution based on convolutional neural networks (CNN) using semantic and syntactic features of sentences. The similarity score between two sentences is computed as follows. First, given a sentence, two matrices are constructed accordingly, which are called the syntax model input matrix and the semantic model input matrix; one records some syntax features, and the other records some semantic features. By experimenting with different arrangements of representing the syntactic and semantic features of the sentences in the matrices, we adopt the most effective way of constructing the matrices. Second, these two matrices are given to two neural networks, which are called the sentence model and the semantic model, respectively. The convolution process of the neural networks of the two models is carried out in multiple perspectives. The outputs of the two models are combined as a vector, which is the representation of the sentence. Third, given the representation vectors of two sentences, the similarity score of these representations is computed by a layer in the CNN. Experiment results show that our algorithm (SSCNN) surpasses the performance MPCPP, which noticeably the best recent work of using CNN for sentence similarity computation. Comparing with MPCNN, the convolution computation in SSCNN is considerably simpler. Based on the results of this work, we suggest that by further utilization of semantic and syntactic features, the performance of sentence similarity measurements has considerable potentials to be improved in the future.     

Influence of luminance and resolution on the perceived quality of black-and-white images on soft displays

Abstract

The perceived quality of an image displayed on a computer monitor depends on a number of different factors that can affect viewers' preferences. Two of these are the luminance of the monitor display and the resolution of the image. The effect of luminance is of interest for applications such as on-line access of images where the computer displays used for viewing these images could have different luminance settings. A relationship between cathode-ray tube (CRT) display luminance and resolution has been shown in previous studies. It was therefore interesting to investigate whether there is a relationship between image resolution and CRT display luminance on perceived image quality. Image resolution is related to the image file size, which is an important factor for applications such as on-line access of images. This work used a CRT display to study the effect of the above-mentioned factors on the perceived quality of the displayed image. Three sets of black-and-white images, each set with a different resolution, were presented to observers at three brightness settings of the computer monitor. Results are discussed regarding the effect of monitor display luminance and image resolution on perceived image quality and the interaction between them. Evaluation of results is further extended to the influence of the different backgrounds of the images. The scene content of the images was also shown to affect the viewers' judgement.     

An Optimum Shift-and-Weighted brightness mapping for low-illumination image restoration

Images captured under low-illumination environments often impose difficulties in revealing objects of interest. An effective approach, Optimum Shift-and-Weighted Brightness Mapping, is here proposed that can optimally enhance the image for higher brightness, information content, and colour vividness. Specifically, the input-output brightness mapping is determined by a shifted spline curve and a larger amplification is allowed for low-brightness pixels. A weighting function is further applied such that high brightness pixels are preserved. The final enhanced image is obtained by inserting the extracted high frequency components from the original input to the brightness boosted image. The algorithm is adaptive to image contents where parameters are optimized using the efficient golden section search instead of relying on user specified coefficients. Experimental results, from a large set of test images, showed that better quality images could be obtained on a variety of low-illumination scenarios as compared to several recent approaches.     

基于多超平面支持向量机的图像语义分类算法

由于图像的低层可视特征与高层语义内容之间存在巨大的语义鸿沟,而基于内容的图像分类和检索准确性极大依赖低层可视特征的描述,本文提出了一种基于多超平面支持向量机的图像语义分类方法.多超平面分类器从优化问题的复杂度和运行泛化能力两方面进行研究,是最优分离超平面分类器一种显而易见的扩展.实验结果表明,本文提出的方法在图像语义分类的准确性方面要优于诸如采用色彩特征和纹理特征的支持向量机分类器的其它方法.     

Image compression by vector quantization with noniterative derivation of a codebook: applications to video and confocal images

We demonstrate an image-compression technique that uses what we believe is a new noniterative codebook generation algorithm for vector quantization. The technique supports rapid decompression and is equally applicable to individual images or to a set of images without the need for interframe processing. Compression with a single-image codebook is tested on (1) ten confocal images of the hindbrain of a mouse embryo, (2) video images of a polystyrene microsphere that is manipulated by a focused laser light, and (3) five fluorescence images of the embryo eye lens taken at different magnifications. The reconstructions are assessed with the normalized mean-squared error and with Linfoot's criteria of fidelity, structural content, and correlation quality. Experimental results with single-image compression show that the technique produces fewer local artifacts than JPEG compression, especially with noisy images. Results with video and confocal image series indicate that single-image codebook generation is sufficient at practical compression ratios for producing acceptable reconstructions for mouse embryo analysis and for viewing optically trapped microspheres. Experiments with the magnified images also reveal that the compression scheme is robust to scaling.