In the article titled “A Multiresolution Spline with Application to Image Mosaics” by Peter J. Burt and Edward H. Adelson, published in 1983, the authors present a novel multiresolution spline technique for combining images into a larger mosaic. This technique addresses the prominent problem in creating photomosaics: differences in grayscale levels at the edges between images, which often make the seams between them visible. To resolve this, they propose a method that ensures a smooth transition without distorting the original images.
In the described procedure, the images to be combined are decomposed into a set of bandpass component filtered images. Then, the component images in each spatial frequency band are assembled into a corresponding bandpass mosaic, integrating through a weighted average within a transition zone proportional to the size of the wavelengths represented in the band. Finally, these bandpass mosaic images are summed to obtain the desired image mosaic. This approach allows the spline to adapt to the scale of elements within the images themselves, facilitating a gradual integration of coarse features near the edges without blurring or degrading finer details in the vicinity of the edge.
A central point of the article is the use of the pyramid structure, which enables highly efficient image processing. Image pyramids are used both for filtering operations and splining. Through pyramid algorithms, the multiresolution spline technique avoids common artifacts in simpler techniques, such as blurred edges or double exposure effects. By decomposing images into frequency bands and carefully recombining them, color and position discrepancies are effectively managed to generate a mosaic without visible seams.
Furthermore, they illustrate the method with various examples, such as combining half of an image of an apple with half of an orange. This example demonstrates how even images with very different features can be integrated without perceptible edges. Similarly, the method was applied to remove block edges in images encoded by block transform, evidencing its usefulness in the visual enhancement of images compressed at low bit rates.
In summary, the multiresolution spline technique described in the article is not only practical and general for forming image mosaics but also offers a unifying framework to perform filtering and splining operations easily and efficiently. This methodology enables the creation of high-quality visual mosaics without common artifacts found in less sophisticated techniques.
