中文摘要：

随着大规模多色巡天项目的完成,测光红移已被视为研究宇宙大尺度结构以及星系形成和演化的有效工具。该文介绍了测光红移的背景、现状、算法及其在天文学中的应用。综述了九种较为常用的估测测光红移的算法,包括HyperZ、颜色-星等-红移关系法（CMR）、多项式回归、基于Kd树的多项式回归、贝叶斯方法、支持向量机（SVMs）、人工神经网络（ANNs）、最近邻或K近邻方法和核回归等。着重讨论和比较了这些算法的效果和性能,同时也对它们的优缺点进行了阐述,并对未来的测光红移算法研究进行了展望。

英文摘要：

With the establishment and development of large digital sky survey projects,astronomic data are measured by TB,even PB,including various photometric and spectroscopic data. Photometric redshifts have shown their superiority compared to spectroscopic ones.So far photometric redshifts have been regarded as an efficient and effective measure for studying the statistical properties of the large-scale structure of the universe and the formation and evolution of galaxies.We illustrate the conception,background and approaches of photometric redshifts, as well as its application in astronomy,then mainly summarize nine approaches to determine photometric redshifts,namely HyperZ,Color-Magnitude-Redshift relation（CMR）,polynomial regression,polynomial regression based on KdTree,Bayesian method,Support Vector Machines （SVMs）,Artificial Neural Networks（ANNs）,K-nearest neighbor and kernel regression.Photometric redshift techniques have been divided into two broad categories：template matching method and empirical training-set method.The former includes HyperZ,and the latter contains CMR,polynomial regression,polynomial regression based on KdTree,Bayesian method,SVMs, ANNs.Another interpolative training-set methods are instance-based learning techniques,which are composed of nearest neighbor,K-nearest neighbor and kernel regression.There are advantages and disadvantages to each approach.Template matching technique relies on fitting model galaxy spectral energy distributions（SEDs）to the photometric data,where the models span a range of expected galaxy redshifts and spectral types.The Achilles heel of the technique is the shortage of large and complete template sets.The training set method depends on representative and complete training sets,moreover it is difficult to extrapolate to regions that are not well sampled by the training set.Unlike the traditional training methods,the best merit of instance-based learning approach is the ability to make predictions with different parameters without needing a retraining