Titlebook: Covariances in Computer Vision and Machine Learning; Hà Quang Minh,Vittorio Murino Book 2018 Springer Nature Switzerland AG 2018

歡樂(lè)中國(guó)

逃避責(zé)任

Aggregate

Ornament

Introduction,eir applications in many disciplines in science and engineering. The practical applications of SPD matrices are numerous, including Diffusion Tensor Imaging (DTI) in brain imaging [5, 29, 66, 95], kernel learning [2, 60] in machine learning, radar signal processing [3, 9, 40], and Brain Computer Interface (BCI) applications [7, 8, 24, 100].

Presbyopia

hankering

Data Representation by Covariance Operatorsis chapter, by employing the feature map viewpoint of kernel methods in machine learning, we generalize covariance matrices to infinite-dimensional covariance operators in RKHS. Since they encode . between input features, they can be employed as a powerful form of data representation, which we explore in subsequent chapters.

吹氣

Geometry of Covariance Operatorsrators. These distances and divergences can then be directly employed in a practical application, e.g., image classification. We emphasize, however, that the concepts we present below are general and applicable in any application involving the comparison of covariance operators.

Alveolar-Bone

maudtin

We then present a statistical interpretation of this framework, which shows that assuming that an image can be represented by a covariance matrix is essentially equivalent to assuming that its features are random variables generated by a multivariate Gaussian probability distribution with mean zero

防銹

d images by covariance matrices, this means that we need to have a similarity measure between covariance matrices. Since covariance matrices, properly regularized if necessary, are symmetric, positive definite (SPD matrices), a natural approach to measuring their similarity is via a distance (or dis