共查询到20条相似文献,搜索用时 0 毫秒
1.
Raja Bala R. Victor Klassen Karen M. Braun 《Journal of the Society for Information Display》2007,15(11):947-957
Abstract— Color characterization is an important step towards achieving accurate color on displays. The characterization process typically uses colorimetric or spectrophotometric instruments to measure displayed colors, and relates these to digital values driving the device. Such measurements can be impractical for consumer applications. This paper presents two techniques for characterizing a display's tone response with no colorimetric or spectrophotometric measurements. The first is a visual technique applicable to devices that exhibit a “gamma” response, such as the cathode‐ray tube. The novelty lies in the replacement of the standard luminance matching with gray‐balancing for the blue channel. This approach substantially reduces observer variation in the gamma estimates for the blue channel. The second technique is applicable for the more general case of devices that do not conform to the gamma response, such as LCDs. The visual task is augmented with a consumer digital camera used as a color‐capture device. The camera tone response is first characterized via a visual task. The characterized camera is then treated as a colorimeter and used to generate a tone‐response characterization for the display. Experiments conducted on projection displays demonstrate that satisfactory quality can be achieved while eliminating the need for costly and tedious measurement. 相似文献
2.
阐述了基于BP神经网络的数码相机特征化方法。采用不同的神经网络结构,建立了数码相机记录的RGB信息和原影像C IEXYZ色度信息之间的非线性对应关系。对NIKON D200数码相机进行了研究,通过实验得到了合理的神经网络结构为3—10—10—3。测试不同的训练样本和测试样本,达到的C IELAB平均色差和最大色差分别为1.9~2.2和6.7~7.4个色差单位。讨论了实验设备的重复性,同时,分析了样本数量对实验结果的影响。实验结果表明:对数码相机的特征化,可采用BP神经网络技术实现较高的精度。 相似文献
3.
Daniel Peralta Isaac Triguero Salvador García Yvan Saeys Jose M. Benitez Francisco Herrera 《国际智能系统杂志》2018,33(1):213-230
Fingerprint classification is one of the most common approaches to accelerate the identification in large databases of fingerprints. Fingerprints are grouped into disjoint classes, so that an input fingerprint is compared only with those belonging to the predicted class, reducing the penetration rate of the search. The classification procedure usually starts by the extraction of features from the fingerprint image, frequently based on visual characteristics. In this work, we propose an approach to fingerprint classification using convolutional neural networks, which avoid the necessity of an explicit feature extraction process by incorporating the image processing within the training of the classifier. Furthermore, such an approach is able to predict a class even for low‐quality fingerprints that are rejected by commonly used algorithms, such as FingerCode. The study gives special importance to the robustness of the classification for different impressions of the same fingerprint, aiming to minimize the penetration in the database. In our experiments, convolutional neural networks yielded better accuracy and penetration rate than state‐of‐the‐art classifiers based on explicit feature extraction. The tested networks also improved on the runtime, as a result of the joint optimization of both feature extraction and classification. 相似文献
4.
深度神经网络(deep neural networks, DNNs)及其学习算法,作为成功的大数据分析方法,已为学术界和工业界所熟知.与传统方法相比,深度学习方法以数据驱动、能自动地从数据中提取特征(知识),对于分析非结构化、模式不明多变、跨领域的大数据具有显著优势.目前,在大数据分析中使用的深度神经网络主要是前馈神经网络(feedforward neural networks, FNNs),这种网络擅长提取静态数据的相关关系,适用于基于分类的数据应用场景.但是受到自身结构本质的限制,它提取数据时序特征的能力有限.无限深度神经网络(infinite deep neural networks)是一种具有反馈连接的回复式神经网络(recurrent neural networks, RNNs),本质上是一个动力学系统,网络状态随时间演化是这种网络的本质属性,它耦合了“时间参数”,更加适用于提取数据的时序特征,从而进行大数据的预测.将这种网络的反馈结构在时间维度展开,随着时间的运行,这种网络可以“无限深”,故称之为无限深度神经网络.重点介绍这种网络的拓扑结构和若干学习算法及其在语音识别和图像理解领域的成功实例. 相似文献
5.
提出了一种新的深度残差网络的拓展模块,有效提高了学习表示的鲁棒性.所提出的方法是一个简单的即插即用模块,即组卷积式编码-解码结构,它可以作为一个额外的信息过滤部件集成到原来的深度残差网络中.利用编码器的下采样来产生信息压缩过的特征图,解码器模块被驱动以产生激活准确的特征图,其能够突出显示输入图片中最具有判别力的区域,最... 相似文献
6.
深度神经网络通常是过参数化的,并且深度学习模型存在严重冗余,这导致了计算和存储的巨大浪费.针对这个问题,本文提出了一种基于改进聚类的方法来对深度神经网络进行压缩.首先通过剪枝策略对正常训练后的网络进行修剪,然后通过K-Means++聚类得到每层权重的聚类中心从而实现权值共享,最后进行各层权重的量化.本文在LeNet,AlexNet和VGG-16上分别进行了实验,提出的方法最终将深度神经网络整体压缩了30到40倍,并且没有精度损失.实验结果表明通过基于改进聚类的压缩方法,深度神经网络在不损失精度的条件下实现了有效压缩,这使得深度网络在移动端的部署成为了可能. 相似文献
7.
PTA工业生产过程中4-CBA的含量是评价其产品质量的重要依据。将深度置信网络和已有的浅层算法相结合,提出基于深度置信网络的4-CBA软测量模型。深度置信网络是一种典型的深度学习算法,该算法在特征学习方面优势显著。根据实验结果,基于深度置信网络的软测量模型能够很好地估计4-CBA含量,和单纯的BP神经网络模型相比,基于深度置信网络的模型预测精度更高。 相似文献
8.
Alexander Heye 《Concurrency and Computation》2019,31(16)
Deep learning has proven itself to be a difficult problem in the HPC space. Although the algorithm can scale very efficiently with a sufficiently large batchsize, the efficacy of training tends to decrease as the batchsize grows. Scaling the training of a single model may be effective in narrow fields such as image classification, but more generalizable options can be achieved when considering alternate methods of parallelism and the larger workflow surrounding neural network training. Hyperparameter optimization, data set segmentation, hierarchical fine tuning, and model parallelism can all provide significant scaling capacity without increasing batchsize and can be paired with a traditional, single‐model scaling approach for a multiplicative scaling improvement. This paper intends to further define and examine these scaling techniques in how they perform individually and how combining them can provide significant improvements in overall training times. 相似文献
9.
Malware has considerably increased recently, posing a serious security danger to both people and enterprises. In order to distinguish and stop the negative effects of malware, a variety of machine and deep learning approaches have been used to detect it. However, while extracting malware features, the feature-to-feature spatial hierarchy is not taken into account by the existing techniques and as a result, information is lost during the pooling operation. Hence, a modified capsule deep neural network was developed in which discriminative features are extracted from three channel image derived from malware binary with considering feature-to-feature spatial hierarchy. Also, conventional capsule deep neural network is modified by adding a global average pooling layer before fully connected layer thereby classified the dataset as malicious or benign without any loss of information. Moreover, these malwares were not accurately classified based on their families using existing variants of convolutional neural network (CNN) since malware family variants can modify due to minute changes in malware binaries. Hence, a hybrid deep convolutional neural network (DCNN) and long-short-term memory (LSTM) has been utilized that determine minute changes in malware binaries using LSTM without vanishing gradient issue and effectively perform malware family classification using DCNN. As a result, the proposed approach successfully identifies malware in executable files and categorizes malware into families with 98.5% accuracy. 相似文献
10.
V. Bhatia M. Hempstead J. Grochocinski N. Sekiguchi A. Okada D. Loeber 《Journal of the Society for Information Display》2009,17(1):47-52
Abstract— Efficient and very‐compact projectors embedded into mobile consumer‐electronic devices, such as handsets, media players, gaming consoles, and GPS units, will enable new consumer use and industry business models. A keystone component for such projectors is a green laser that is commensurately efficient and compact. A synthetic green‐laser architecture is described that can achieve efficiencies of 15%. The architecture consists of an infrared distributed Bragg reflector laser coupled into a second‐harmonic‐generation device for conversion to green. 相似文献
11.
Abstract— A major issue when setting up multi‐projector tiled displays is the spatial non‐uniformity of the color throughout the display's area. Indeed, the chromatic properties do not only vary between two different projectors, but also between different spatial locations inside the displaying area of one single projector. A new method for calibrating the colors of a tiled display is presented. An iterative algorithm to construct a correction table which makes the luminance uniform over the projected area of one single projector is presented first. This so‐called intra‐projector calibration uses a standard camera as a luminance measuring device and can be processed in parallel for all projectors. Once the color inside each projector is spatially uniform, the set of displayable colors — the color gamut — of each projector is measured. On the basis of these measurements, the goal of the inter‐projector calibration is to find an optimal gamut shared by all the projectors. Finding the optimal color gamut displayable by n projectors in time O(n) is shown, and the color conversion from one specific color gamut to the common global gamut is derived. The method of testing it on a tiled display consisting of 48 projectors with large chrominance shifts was experimentally validated. 相似文献
12.
Abstract— The advent of affordable direct‐diode lasers changes all the rules for optical designs and the associated technologies that generate the images from laser light. These new lasers are forseen as driving fundamental changes in the size, power consumption, cost, resolution, and even the uses for pico‐projectors. This paper discusses these topics from the perspective of laser‐light‐illuminated LCOS microdisplays. 相似文献
13.
深度神经网络模型压缩综述 总被引:1,自引:0,他引:1
近年来;随着深度学习的飞速发展;深度神经网络受到了越来越多的关注;在许多应用领域取得了显著效果。通常;在较高的计算量下;深度神经网络的学习能力随着网络层深度的增加而不断提高;因此深度神经网络在大型数据集上的表现非常卓越。然而;由于其计算量大、存储成本高、模型复杂等特性;使得深度学习无法有效地应用于轻量级移动便携设备。因此;压缩、优化深度学习模型成为目前研究的热点。当前主要的模型压缩方法有模型裁剪、轻量级网络设计、知识蒸馏、量化、体系结构搜索等。对以上方法的性能、优缺点和最新研究成果进行了分析总结;并对未来研究方向进行了展望。 相似文献
14.
应用灰度图像增强方法对真彩图像进行增强,往往都会产生色彩偏离,影响增强结果和视觉效果。因此基于人眼视觉系统对亮度变化比较敏感,提出在HSV色彩空间,应用PCNN模型进行增强的方法。将真彩图像由RGB空间变换到HSV空间,保持色度H和饱和度S不变,结合入射反射模型,利用脉冲耦合神经网络(PCNN),对亮度V通道进行增强处理。将HSV空间得到的增强图像变换到RGB空间。实验证实,对一些对比度低、细节不明显的图像应用此方法进行增强,色彩基本无偏差,细节部分明晰,动态范围压缩较好,视觉效果得到了较大的改善。 相似文献
15.
Li-Min Fu 《Applied Intelligence》1992,2(1):93-103
A novel approach to rule refinement based upon connectionism is presented. This approach is capable of performing rule deletion, rule addition, changing rule quality, and modification of rule strengths. The fundamental algorithm is referred to as the Consistent-Shift algorithm. Its basis for identifying incorrect connections is that incorrect connections will often undergo larger inconsistent weight shift that correct ones during training with correct samples. By properly adjusting the detection threshold, incorrect connections would be uncovered, which can then be deleted or modified. Deletion of incorrect connections and addition of correct connections then translate into various forms of rule refinement just mentioned. The viability of this approach is demonstrated empirically. 相似文献
16.
Abstract— In the present set of experiments, we examined the mechanisms underlying color break‐up (CBU), a phenomenon observed when images produced with a color‐sequential projector are viewed. The perceived position of CBU was measured during fast eye movement, saccade with static and briefly flashed stimuli. Results showed that CBU did not simply correspond to the locus of the stimulus on the retina during saccades, because the width of the CBU perception was narrower than the distance of the eye movements. This effect was thought to be related to visual stability, which allows objects to be perceived as stationary even when the eyes move and the retinal image changes position. Visual stability is assumed to operate by compensating for the change in retinal image position using eye‐position signals; however, this compensation is imperfect during saccades. Thus, incomplete compensation results in a CBU perception that is of a narrower width than the amplitude of the saccade. In conclusion, CBU cannot be simulated with moving video cameras because it results largely from the mechanisms of visual perception. Large inter‐individual differences in perception of CBU were also found. This observation also supports the idea that CBU depends on the mechanism of perception. 相似文献
17.
随着计算机和社交网络的飞速发展, 图像美感的自动评价产生了越来越大的需求并受到了广泛关注. 由于图像美感评价的主观性和复杂性, 传统的手工特征和局部特征方法难以全面表征图像的美感特点, 并准确量化或建模. 本文提出一种并行深度卷积神经网络的图像美感分类方法, 从同一图像的不同角度出发, 利用深度学习网络自动完成特征学习, 得到更为全面的图像美感特征描述; 然后利用支持向量机训练特征并建立分类器, 实现图像美感分类. 通过在两个主流的图像美感数据库上的实验显示, 本文方法与目前已有的其他算法对比, 获得了更好的分类准确率. 相似文献
18.
19.
借鉴数码相机检校的方法对投影器进行检校,即采用的投影器检校模型与数码相机的检校模型类似,但检校精度却低于数码相机的检校精度,从物方详细分析其原因并对由于纸张(避免成像在平面格网板上虚拟影像的格网点与真实格网点的混淆,使用白纸进行遮挡)厚度引起的物方点误差,通过模拟数据对物方点的Z值进行补偿以提高投影器检校精度。经过实验及结果分析,证明该方法具有很好的可行性。 相似文献
20.
Rodney L. Heckaman Mark D. Fairchild 《Journal of the Society for Information Display》2006,14(9):755-761
Abstract— The effect of white‐channel enhancement as implemented in the Texas Instrument DLP? digital projector technology is evaluated theoretically using both the CIELAB and the CIECAM02 color appearance models and experimentally through psychophysical testing using real images. Both theory and test results confirm a compression of perceptual gamut in both chroma and colorfulness as a result of the added white channel. Hence, while this technology is ideal for viewing graphics and text under ambient conference‐room conditions where lightness contrast is important, it is necessarily less than ideal for viewing images or in home‐theater environments where color is important. 相似文献