基于受损区域运动特征的时域错误隐藏算法

为解决视频压缩数据在网络传输中的误码或丢包问题,提出了一种适用于数字视频编码标准H.264的时域错误隐藏算法。该算法结合H.264中多尺寸编码块模式的特点,通过对受损块邻域块的模式分析来判定受损块模式,并针对各子块的运动类型分别建立相应的运动矢量候选集,从而选择最优运动矢量重建受损块。实验结果表明,该算法有效增强了对复杂运动区域重建的适应性,在不同的网络丢包率环境下均可实现较好的主观和客观重建图像质量。     

结构相似度的立体视频错误隐藏

针对立体视频传输中右视点整帧丢失,提出了一种基于结构相似度(Structural Similarity,SSIM)的立体视频右视点整帧丢失错误隐藏算法。首先,提出了时域SSIM和视间SSIM的概念。然后,根据视频序列的时域相关性,将前一时刻右视点图像宏块的预测方式作为丢失图像宏块的预测方式。接着,将前一时刻右视点图像以宏块为单位进行时域和视间匹配,求取其以像素为单位的时域SSIM映射图和视间SSIM映射图。最后,计算并比较前一时刻右视点图像每个宏块的时域SSIM和视间SSIM值,得到每个宏块的预测方式,将其预测方式作为丢失帧中宏块的预测方式,从而使用运动补偿预测或者视差补偿预测的方法进行恢复。实验结果表明,与传统的算法和Pang的算法相比,PSNR值分别提高了2.76 dB和3.43 dB,且本文算法主观效果较好。     

视频通信中的抗误码方法研究

提出了非等重前向纠错，时域／空域误码掩盖，交互式防误码扩散和主观质量控制策略等多种新颖，实用的抗误码方法，实验结果表明，在高误码环境下的视频通信中，综合采用以上的抗误码方法，有效地提高了视频码流的抗误码能力，恢复视频的主客观质量得到较大的改善。     

面向HEVC的深度视频预处理

通过深度相机或者深度估计软件获取的深度视频准确性差,影响深度视频的编码压缩性能。因此,本文提出了一种面向HEVC的深度视频预处理算法。首先提取深度视频的边缘获取边缘掩膜;其次根据边缘掩膜对深度序列分类进行空域平滑;再次将彩色和空域平滑后的深度视频进行时空转换,并以转换后的彩色视频为辅助对转换后的深度视频进行时域平滑,以增强深度视频的时间相关性;最后对时域平滑处理完的深度视频进行逆时空转换,获取最终的预处理深度视频。实验结果表明,提出的处理方法既可以保证虚拟视点质量基本不变,又能够节省7.00%～23.78%的码率。     

基于运动一致性的视频对象分割方法研究

视频运动对象检测和分割是图像处理中最具挑战性的问题之一。针对目前大部分分割算法相当复杂而且计算量大的问题，提出了一种基于运动一致性的视频对象分割方法。该方法从MPEG压缩码流中提取运动矢量场来分割视频对象，首先对运动矢量场进行滤波和校正，然后进行全局运动补偿得到对象的绝对运动矢量场，最后采用K-means聚类算法对运动矢量场进行聚类分析从而分割出感兴趣的视频运动对象。MPEG标准测试序列的试验结果证明，该方法是有效的。     

根据帧间运动特性的变尺寸宏块运动估计

提出了一种新的用于视频编码的变尺寸宏块运动估计方法。该方法通过粗尺度 16×16 的当前块和邻域块运动矢量之间的差值 S 判断场景运动的复杂度,再根据差值 S 和当前矢量,决定宏块尺寸。场景运动平稳的部分采用 16×16 的宏块,复杂的部分采用 8×8 或 4×4 的宏块。实验表明,低码率条件下,采用普通运动补偿算法,该变尺寸宏块运动估计方法的预测帧的平均 PSNR 值比常规的固定尺寸块运动估计约提高 1.1-2.7 dB,并显著降低了方块效应。     

改进的空域平滑的边信息插值算法

针对分布式视频编码(DVC)中的边信息重建问题,提出了一种空域平滑的边信息重建算法.该算法在关键帧之间的前向运动估计时,设置代价函数阈值保证目标运动轨迹的线性连续.对于运动平稳区域,采用双向运动补偿插值生成边信息;对于其它区域,在关键帧中确定多个候选块,并选择与待插值块的4个相邻块具有最小边界匹配误差的候选块作为边信息...     

图像处理中的误差隐藏问题研究

视频信息必须经过压缩以便在低带宽信道上传输,而压缩数字媒体对无线信道误码十分敏感,容易发生传输错误,误差隐藏技术是弥补此类问题的方法之一。本文在对现有的误差隐藏处理方法分析的基础上,提出了一种基于模式选择的时空域误差隐藏技术。此方法不仅从时、空两方面考虑,分别应用时域取代法和加权内插空域法。还通过边界检测,区分了平坦型和边界穿越型两类宏块,减少算法复杂性。仿真实验表明,该方法可有效地抑制视频差错的扩散,取得较好的差错掩盖效果。     

视频编码中基于时空信息运动估计的研究

提出了一种基于视频图像的时间和空间相关性信息的运动矢量预测方法，由于相领两帧图像中的内容应具有相关性，图像中物体的运动轨迹也具有相似性，因此可以利用前一帧图像中宏块的运动矢量确定运动轨迹，进而估计当前宏块的运动矢量，但是这种运动估计是受限的，因为估计运动矢量时的宏块顺序是逐行的按先后顺充进行，在当前帧中，当前宏块预测时只能利用具有因果关系的相邻宏块，但其与相邻宏块并非都具有因果关系，因此给出了预测矢量的资格检验方法，用以判断和剔除伪预测矢量，在搜索过程中采用菱形搜索（Pyramid)，相邻的两次搜索过程，要去除两个重叠区域，上述算法减少计算量，对Suzie序列测试结果，平均搜索次数是14－17次，并且解码图像的信噪比PSNR与FS搜索相比降级不多。     

一种结合时间轴滤波的视频消噪新方法

为了提高含噪视频图像的质量,提出了一种二维小波域自适应滤波与时域时间轴滤波相结合的视频图像消噪新方法。首先,对视频序列的各帧在二维小波域中进行自适应滤波,之后在时域中进行时间轴滤波。对于二维小波域滤波算法,提出了一种高效的自适应阈值选取方案;时间轴滤波器则是结合了运动检测和递归平均。实验结果表明,其消噪效果要优于单纯的二维小波域滤波方法。     

Heterodyne temporal speckle-pattern interferometry

In temporal speckle-pattern interferometry deformation information is extracted by a Fourier transform technique from the speckle pattern that is recorded over a period of time as the object is deformed. A limitation of the experimental arrangements reported to date is that the direction of the deformation cannot be determined. We propose removing this limitation by using the heterodyne principle. Some experimental results that were obtained by use of a rotating half-wave-plate frequency shifter are presented.     

Compact optical temporal processors

Optical signal processing can be done with time-lens devices. A temporal processor based on chirp-z transformers is suggested. This configuration is more compact than a conventional 4-? temporal processor. On the basis of implementation aspects of such a temporal processor, we did a performance analysis. This analysis leads to the conclusion that an ultrafast optical temporal processor can be implemented.     

Integrating cavities: temporal response

The temporal response of an integrating cavity is examined and compared with the results of a Monte Carlo analysis. An important parameter in the temporal response is the average distance d between successive reflections at the cavity wall; d was calculated for several specific cavity designs--spherical shell, cube, right circular cylinder, irregular tetrahedron, and prism; however, only the calculation for the spherical shell and the right circular cylinder will be presented. A completely general formulation of d for arbitrary cavity shapes is then derived, d =4V/S where V is the volume of the cavity, and S is the surface area of the cavity. Finally, we consider an arbitrary cavity shape for which each flat face is tangent to a single inscribed sphere of diameter D (a curved surface is considered to be an infinite number of flat surfaces). We will prove that for such a cavity d =2D/3, exactly the same as d for the inscribed sphere.     

Visual adjustments to temporal blur

After observers have adapted to an edge that is spatially blurred or sharpened, a focused edge appears too sharp or blurred, respectively. These adjustments to blur may play an important role in calibrating spatial sensitivity. We examined whether similar adjustments influence the perception of temporal edges, by measuring the appearance of a step change in the luminance of a uniform field after adapting to blurred or sharpened transitions. Stimuli were square-wave alternations (at 1 to 8 Hz) filtered by changing the slope of the amplitude spectrum. A two-alternative-forced-choice task was used to adjust the slope until it appeared as a step change, or until it matched the perceived transitions in a reference stimulus. Observers could accurately set the waveform to a square wave, but only at the slower alternation rates. However, these settings were strongly biased by prior adaptation to filtered stimuli, or when the stimuli were viewed within temporally filtered surrounds. Control experiments suggest that the latter induction effects result directly from the temporal blur and are not simply a consequence of brightness induction in the fields. These results suggest that adaptation and induction adjust visual coding so that images are focused not only in space but also in time.     

Processing of second-order motion stimuli in primate middle temporal area and medial superior temporal area

Two rhesus monkeys were subjects in a direction-discrimination task involving moving stimuli defined by either first- or second-order motion. Two different second-order motion stimuli were used: drift-balanced motion consisting of a rectangular field of stationary dots and theta motion consisting of the same rectangular field with dots moving in the direction opposite to that of the object. The two types of stimuli involved different segmentation cues between the moving object and the background: temporal structure of the luminance (flicker) in the case of drift-balanced motion and opposed motion in the case of the theta-motion stimulus. Our monkeys were able to correctly report the direction of each stimulus. Single-unit recordings from the middle temporal (MT) and medial superior temporal (MST) areas revealed that 16 out of 38 neurons (41%) from area MT and 34 out of 68 neurons (50%) from area MST responded in a directionally selective manner to the drift-balanced stimulus. The movement of an object defined by theta motion is not explicitly encoded in the neuronal activity in areas MT or MST. Our results do not support the hypothesis that the neuronal activity in these areas codes for the direction of stimulus movement independent of specific stimulus parameters. Furthermore, our results emphasize the relevance of different segmentation cues between figure and background. Therefore the notion that there are multiple sites responsible for the processing of second-order motion is strongly supported.     

A survey of temporal data mining

Data mining is concerned with analysing large volumes of (often unstructured) data to automatically discover interesting regularities or relationships which in turn lead to better understanding of the underlying processes. The field of temporal data mining is concerned with such analysis in the case of ordered data streams with temporal interdependencies. Over the last decade many interesting techniques of temporal data mining were proposed and shown to be useful in many applications. Since temporal data mining brings together techniques from different fields such as statistics, machine learning and databases, the literature is scattered among many different sources. In this article, we present an overview of techniques of temporal data mining. We mainly concentrate on algorithms for pattern discovery in sequential data streams. We also describe some recent results regarding statistical analysis of pattern discovery methods.     

Extended-range temporal electronic speckle pattern interferometry

In recent years the availability of high-speed digital video cameras has motivated the study of electronic speckle pattern interferometry (ESPI) in the time domain. To this end a properly sampled temporal sequence of N-fringe patterns is used to analyze the temporal experiment. Samples of temporal speckle images must fulfill the Nyquist criteria over the time axis. When the transient phenomena under study are too fast, the required sampling frequency over time may not be fulfilled. In that case one needs to extend the measuring range of the algorithm used to extract the modulating phase. We analyze how to use short laser pulses or short video acquisition times with fairly long temporal separation among them to estimate the modulating phase of a dynamic ESPI experiment. The only requirement is that the modulating phase being estimated be properly sampled in the spatial domain.     

Dynamic coding of temporal luminance variation

The range of variation in environmental stimuli is much larger than the visual system can represent. It is therefore sensible for the system to adjust its responses to the momentary input statistics of the environment, such as when our pupils contract to limit the light entering the eye. Previous evidence indicates that the visual system increasingly centers responses on the mean of the visual input and scales responses to its variation during adaptation. To what degree does adaptation to a stimulus varying in luminance over time result in such adjustment of responses? The first two experiments were designed to test whether sensitivity to changes in the amplitude and the mean of a 9.6° central patch varying sinusoidally in luminance at 0.6 Hz would increase or decrease with adaptation. This was also tested for a dynamic peripheral stimulus (random patches rotating on the screen) to test to what extent the effects uncovered in the first two experiments reflect retinotopic mechanisms. Sensitivity to changes in mean and amplitude of the temporal luminance variation increased sharply the longer the adaptation to the variation, both for the large patch and the peripheral patches. Adaptation to luminance variation leads to increased sensitivity to temporal luminance variation for both central and peripheral presentation, the latter result ruling retinotopic mechanisms out as sole explanations for the adaptation effects.