Correlation Filters Theory


Support vector machine (SVM) classifiers are popular in many computer vision tasks. In most of them, the SVM classifier assumes that the object to be classified is centered in the query image which might not always be valid, e.g., when locating and classifying a particular class of vehicles in a large scene. In this paper we introduce a new classifier called Maximum Margin Correlation Filter (MMCF), which while exhibiting the good generalization capabilities of SVM classifiers is also capable of localizing objects of interest, thereby avoiding the need for image centering as is usually required in SVM classifiers. In other words, MMCF can simultaneously localize and classify objects of interest. We test the efficacy of the proposed classifier on three different tasks: vehicle recognition, eye localization, and face classification. We demonstrate that MMCF outperforms SVM classifiers and also well-known correlation filters.


alt text

Application of a correlation filter to a query image.

alt text

A Mind Map of various Linear Correlation Filter designs proposed in the literature.


  • Introduced a new correlation filter design for combing the localization properties of correlation filters and the margin maximizing properties of SVMs [3,8].
  • Introduced multi-channel correlation filters for object part localization with applications to object alignment [4].
  • Introduced correlation filter designs for predicting structured outputs (binary bit information) with applications to biometric key-binding [7].
  • Introduced a method for learning a discriminatively reduced ensemble of exemplar correlation filters with applications to 3D pose estimation of vehicles.
  • Historically correlation filter designs formulated in the Fourier domain suffer from aliasing effects due to circular correlation. We introduce aliasing free for all previously proposed correlation filter designs [1].