The paper by J. P. Lewis from Industrial Light & Magic discusses the efficient computation of normalized cross-correlation (NCC) in the transform domain, which is crucial for feature matching applications. Traditional methods often compute NCC in the spatial domain due to the lack of a simple frequency domain expression for NCC. However, this paper introduces a method to efficiently normalize unnormalized cross-correlation using precomputed integrals of the image and image squared over the search window. The introduction highlights the importance of cross-correlation in feature detection and matching, noting that while it can be efficiently computed in the frequency domain using the fast Fourier transform, the normalized form (NCC) does not have a straightforward frequency domain expression. This has led to the use of spatial domain methods, which are computationally expensive. The paper then reviews various feature tracking approaches, including the Sequential Similarity Detection Algorithm (SSDA), gradient descent search, snakes (active contour models), and multi-resolution techniques. It concludes that NCC remains a viable choice for many applications despite its limitations in handling scale, rotation, and perspective distortions. The main contribution of the paper is a method to compute NCC efficiently in the transform domain. By precomputing the mean and energy of the image and feature, the complexity of the computation is significantly reduced. This approach is particularly beneficial when the search window is large compared to the feature size, making the transform domain method more efficient. Performance results are presented, showing that the fast NCC algorithm can reduce the time required for high-resolution feature tracking from an overnight process to a lunch break. The algorithm has been successfully used in special effects image processing, such as in the movie "Forest Gump," and has been compared with commercial image compositing systems, demonstrating its effectiveness and efficiency.The paper by J. P. Lewis from Industrial Light & Magic discusses the efficient computation of normalized cross-correlation (NCC) in the transform domain, which is crucial for feature matching applications. Traditional methods often compute NCC in the spatial domain due to the lack of a simple frequency domain expression for NCC. However, this paper introduces a method to efficiently normalize unnormalized cross-correlation using precomputed integrals of the image and image squared over the search window. The introduction highlights the importance of cross-correlation in feature detection and matching, noting that while it can be efficiently computed in the frequency domain using the fast Fourier transform, the normalized form (NCC) does not have a straightforward frequency domain expression. This has led to the use of spatial domain methods, which are computationally expensive. The paper then reviews various feature tracking approaches, including the Sequential Similarity Detection Algorithm (SSDA), gradient descent search, snakes (active contour models), and multi-resolution techniques. It concludes that NCC remains a viable choice for many applications despite its limitations in handling scale, rotation, and perspective distortions. The main contribution of the paper is a method to compute NCC efficiently in the transform domain. By precomputing the mean and energy of the image and feature, the complexity of the computation is significantly reduced. This approach is particularly beneficial when the search window is large compared to the feature size, making the transform domain method more efficient. Performance results are presented, showing that the fast NCC algorithm can reduce the time required for high-resolution feature tracking from an overnight process to a lunch break. The algorithm has been successfully used in special effects image processing, such as in the movie "Forest Gump," and has been compared with commercial image compositing systems, demonstrating its effectiveness and efficiency.