This paper presents a technique for computing 2D component velocity from image sequences. The method uses a family of spatiotemporal velocity-tuned linear filters to represent the image sequence. Component velocity is derived from the local first-order behavior of surfaces of constant phase. The technique is linear, efficient, and suitable for parallel processing. It is local in space-time, robust to noise, and allows multiple estimates within a single neighborhood. Promising results are reported from experiments with realistic image sequences, including cases with significant perspective deformation. The paper addresses the quantitative measurement of velocity in image sequences, focusing on accuracy, robustness to smooth contrast variations and affine deformation, localization in space-time, noise robustness, and the ability to discern different velocities within a single neighborhood. The approach is based on phase information in a local-frequency representation of the image sequence, produced by a family of velocity-tuned linear filters. The velocity measurements are limited to component velocity, which is the projected component of 2D velocity onto directions normal to oriented structure in the image. The combination of these measurements to derive the full 2D velocity is briefly discussed. The paper discusses the advantages of phase-based methods, including velocity resolution, subpixel accuracy, and robustness to smooth contrast changes and affine deformations. The three main advantages of phase-based methods are: (1) velocity resolution, (2) subpixel accuracy, and (3) robustness. The paper outlines the initial image representation, defines component velocity in terms of local phase behavior, and outlines a method for its measurement. The technique has been implemented, and several demonstrations of the accuracy and robustness of the technique are given. These experiments involve real and synthetic image sequences with significant time-varying perspective distortions.This paper presents a technique for computing 2D component velocity from image sequences. The method uses a family of spatiotemporal velocity-tuned linear filters to represent the image sequence. Component velocity is derived from the local first-order behavior of surfaces of constant phase. The technique is linear, efficient, and suitable for parallel processing. It is local in space-time, robust to noise, and allows multiple estimates within a single neighborhood. Promising results are reported from experiments with realistic image sequences, including cases with significant perspective deformation. The paper addresses the quantitative measurement of velocity in image sequences, focusing on accuracy, robustness to smooth contrast variations and affine deformation, localization in space-time, noise robustness, and the ability to discern different velocities within a single neighborhood. The approach is based on phase information in a local-frequency representation of the image sequence, produced by a family of velocity-tuned linear filters. The velocity measurements are limited to component velocity, which is the projected component of 2D velocity onto directions normal to oriented structure in the image. The combination of these measurements to derive the full 2D velocity is briefly discussed. The paper discusses the advantages of phase-based methods, including velocity resolution, subpixel accuracy, and robustness to smooth contrast changes and affine deformations. The three main advantages of phase-based methods are: (1) velocity resolution, (2) subpixel accuracy, and (3) robustness. The paper outlines the initial image representation, defines component velocity in terms of local phase behavior, and outlines a method for its measurement. The technique has been implemented, and several demonstrations of the accuracy and robustness of the technique are given. These experiments involve real and synthetic image sequences with significant time-varying perspective distortions.