Most signals you get by sampling real world objects will be fairly sparse. You can expect anywhere between 50% and 95% of the coefficients in a wavelet or fouri

Most signals you get by sampling real world objects will be fairly sparse. You can expect anywhere between 50% and 95% of the coefficients in a wavelet or fourier domain to be near zero depending on the class of signal. However, be wary of algorithms that expect K as input, as they are asking you to classify your signal before analysis.