Comparison of Image Interpolation Methods Applied to Least Squares Matching

Least squares matching requires appropriate interpolation of the gray values in the search window corresponding to a template. This paper reports an experiment conducted to evaluate image interpolation methods on matching accuracy of least squares matching by using 54 diverse images. Three popular methods in photogrammetry: bilinear interpolation [BL], cubic convolution by Riffman [C1], and cubic convolution by Simon [C2] were investigated. The results demonstrate that [C2] can produce better matching results than [BL] and [C1] in most cases when an image has no noise or smaller noises. Meanwhile, the results demonstrate that there is nothing to choose among three interpolation methods when an image has larger noises. Since the differences of the matching accuracy among three methods were small enough to be neglected, we conclude that [BL] would be the best interpolation method applied to least squares matching in most cases considering its inexpensive computational cost.