Effects of lossy wavelet-based compression on classification accuracy
Brian M. Schmanske, M. Loew · 2002
The effect of 2D wavelet-based compression on overall classification accuracy is examined using simulated univariate images containing two classes of objects. Compressed images are analyzed using a maximum-likelihood estimator. Errors are separated into "pure" and "mixed" pixel categories. The study considers images of varying scale. The lowpass images resulting from wavelet decomposition showed similar results observed by decreasing spatial resolution. Specifically, a valley may occur in the relationship between classification error and compression ratio.