Considering that the power spectral density of 1/f noise signal varies inversely with frequency and sparse decomposition is able to create dictionaries according to signals' own features, in this paper we propose a new method to estimate the parameter of 1/f noise of high power semiconductor laser diode based on sparse decomposition, and create a dictionary whose feature is similar to 1/f noise. Sparse decomposition of white noise and 1/f noise aliased signal is implemented by Matching-Pursuit(MP)algorithm. The experimenat results indicate that sparse decomposition could effectively estimate the parameter of 1/f noise signal covering white noise and the estimated parameter is in accordance with the measured result of spectrum analyzer. Finally, the superiority of the dictionary created in this paper is illustrated by comparing with a different dictionary.