计算机英语论文(中英双语).doc

稀疏表示计算机视觉和模式识别

从抽象技术的现象已经可以开始看到稀疏信号在电脑视觉产生重大影响，通常在非传统的应用场合的目标不仅是要获得一个紧凑的高保真度表示的观察信号，而且要提取语义信息。非常规词典在字典的选择中扮演了重要的角色，衔接的差距或学习、训练样本同来获得自己提供钥匙，解出结果和附加语义意义信号稀疏表示。理解这种非传统的良好性能要求词典把新的算法和分析技术。本文强调了一些典型例子：稀疏信号的表现如何互动的和扩展计算机视觉领域，并提出了许多未解的问题为了进一步研究。
稀疏表现已经被证明具有非常强大的工具，获取、表示、压缩高维信号的功能。它的成功主要是基于这个事实，即重要类型的信号如声音和图像，稀疏表示很自然地就固定基地或串连这样的基地。此外，高效、大概有效算法说明基于凸优化一书提供了计算这样的陈述。
虽然这些应用在经典信号处理的铺垫下,已经可以在电脑视觉上形成一个我们经常更感兴趣的内容或语义，而不是一种紧凑、高保真的表示。一个人可能会理所当然地知道是否可以有用稀疏表示为视觉任务。答案很大程度上是积极的：在过去的几年里，变化和延伸的最小化已应用于许多视觉任务。
稀疏表示的能力是揭示出语义信息，大部分来自于一个简单但重要的性质数据：虽然照片所展示的图像是在非常高自然的空间，在许多同类应用中图像属于次级结构。也就是说他们在接近低维子空间或层次。如果发现一个收集的样本分布，我们理应期望一个典型的样品有一个稀疏表示理论的基础。
然而，想要成功地把稀疏表示应用于电脑视觉，我们通常是必须面对的一个额外的问题，如何正确选择依据。这里的数据选择不同于在信号处理的传统设置，基于指定的环境具有良好的性能可以被假定。在电脑视觉方面，我们经常要学习样本图像的任务词典，我们不得不用一个连贯的思想来贯穿工作。因此，我们需要扩展现有的理论和稀疏表示算法新情况。
自动人像识别仍然是最具有挑战性的应用领域和计算机视觉的难题。在理论基础实验上，稀疏表示在近期获得了显著的进展。
该方法的核心是选择一个明智的字典作为代表，用来测试信号稀疏线性组合信号。我们首先要简单的了解令人诧异的人脸识别途径是有效的解决办法。反过来，人脸识别实例在稀疏表示光曝光之前揭示了新的理论现象。
之前稀疏表示的部分用机器检查并且应用，在一个完全词典里组成的语义信息本身产生的样品。对于许多数据不是简单的应用，这是合乎情理的词典，使用一个紧凑的数据得到优化目标函数的一些任务。本节概述学习方法那种词典，以及这些方法应用在计算机视觉和图像处理。
通过近年来我们对稀疏编码和优化的应用的理解和启发，如面部识别一节描述的例子，我们提出通过稀疏数据编码构造，利用它建立了受欢迎的机器学习任务。在一个图的数据推导出研究学报。2009年3月5乘编码每个数据稀疏表示的剩余的样本，并自动选择最为有效的邻居为每个数据。通过minimization稀疏表示的计算自然的性能满足净水剂结构。 此外，我们将会看到描述之间的关系进行了实证minimization线性数据的性能，可以显著提高现有的基于图论学习算法可行性。
摘自：期刊IEEE的论文- PIEEE ，第一卷
英文翻译

SPARSE REPRESENTATION FOR COMPUTER VISION AND PATTERN RECOGNITION

Abstract—Techniques from sparse signal representation are beginning to see significant impact in computer vision, often on non-traditional applications where the goal is not just to obtain a compact high-fidelity representation of the observed signal, but also to extract semantic information. The choice of dictionary plays a key role in bridging this gap: unconventional dictionaries consisting of, or learned from, the training samples themselves provide the key to obtaining state-of-theart results and to attaching semantic meaning to sparse signal representations. Understanding the good performance of such unconventional dictionaries in turn demands new algorithmic and analytical techniques. This review paper highlights a few representative examples of how the interaction between sparse signal representation and computer vision can enrich both fields and raises a number of open questions for further study.

Sparse signal representation has proven to be an extremely powerful tool for acquiring, representing, and compressing high-dimensional signals. This success is mainly due to the fact that important classes of signals such as audio and images have naturally sparse representations with respect to fixed bases, or concatenations of such bases. Moreover, efficient and provably effective algorithms based on convex optimization or greedy pursuit are available for computing such representations with high fidelity.

While these successes in classical signal processing applications are inspiring, in computer vision we are often more interested in the content or semantics of an image rather than a compact, high-fidelity representation. One might justifiably wonder, then, whether sparse representation can be useful at all for vision tasks. The answer has been largely positive: in the past few years, variations and extensions of minimization have been applied to many vision tasks.

The ability of sparse representations to uncover semantic information derives in part from a simple but important property of the data: although the images are naturally very high dimensional, in many applications images belonging to the same class exhibit degenerate structure. That is, they lie on or near low-dimensional subspaces, or stratifications. If a collection of representative samples are found for the distribution, we should expect that a typical sample have a very sparse representation with respect to such a basis.

However, to successfully apply sparse representation to computer vision tasks, we typically have to address the additional problem of how to correctly choose the basis for representing the data. This is different from the conventional setting in signal processing where a given basis with good property (such as being sufficiently incoherent) can be assumed. In computer vision, we often have to learn from given sample images a task-specific (often over complete) dictionary; or we have to work with one that is not necessarily incoherent. As a result, we need to extend the existing theory and algorithms for sparse representation to new scenarios.

Automatic face recognition remains one of the most visible and challenging application domains of computer vision . Foundational results in the theory of sparse representation have recently inspired significant progress on this difficult problem.

The key idea is a judicious choice of dictionary: representing the test signal as a sparse linear combination of the training signals themselves. We will first see how this approach leads to simple and surprisingly effective solutions to face recognition. In turn, the face recognition example reveals new theoretical phenomena in sparse representation that may seem surprising in light of prior results.

The previous sections examined applications in vision and machine learning in which a sparse representation in an over complete dictionary consisting of the samples themselves yielded semantic information. For many applications, however, rather than simply using the data themselves, it is desirable to use a compact dictionary that is obtained from the data by optimizing some task-specific objective function. This section provides an overview of approaches to learning such dictionaries, as well as their applications in computer vision and image processing.

Enlightened by recent advances in our understanding of sparse coding by optimization and in applications such as the face recognition example described in the previous section, we propose to construct the so-called graph via sparse data coding, and then harness it for popular graph-based machine learning tasks. An graph over a dataset is derived PROCEEDINGS OF IEEE, MARCH 2009 5 by encoding each datum as the sparse representation of the remaining samples, and automatically selects the most informative neighbors for each datum. The sparse representation computed by-minimization naturally satisfies the properties of sparsity and adaptivity. Moreover, we will see empirically that characterizing linear relationships between data samples via-minimization can significantly enhance the performance of existing graph-based learning algorithms.