This paper describes a method for computing a blur factor using DWT. Looked pretty straight forward but instead of detecting sharpness it's detecting blurredness. Seems it detects edges first (simple convolution) and then uses DWT to accumulate and score it.