The spectral parameters of many substances within the terahertz band have fingerprint characteristics, which is the basis of the application of the terahertz technology in many fields such as security inspection. However, it is required that the upper and lower surfaces of samples should be parallel and smooth for the Duvillaret algorithm, which is commonly used to extract the optical parameters of materials in terahertz time-domain spectroscopy (THz-TDS). Moreover, surface roughness is inevitable, and laboratory sample preparation methods such as mold pressing can not be utilized to ensure the parallelism and smoothness of the surface in many potential practical applications, especially for solid samples. Therefore, a universal optical parameter extraction algorithm that is not easily affected by the surface roughness of samples is needed to increase the application ability of THz-TDS in different situations. In this paper, the optical path of the transmission of the terahertz wave through rough samples is analyzed, and a terahertz transmission model of rough samples, including roughness, refractive index and extinction coefficient parameters, is established. Under the condition of known roughness and average thickness, the applicability of the transmission model is verified by extracting the refractive index and absorption coefficient of rough samples with characteristic absorption peaks in the terahertz band. -----The α-Lactose and L-Histidine which have typical absorption features at terahertz frequencies are selected as experimental material. Pellets at the same concentration are prepared and further separated into the rough surface group and smooth surface group. The refractive index, as well as absorption coefficient, are calculated with both our proposed model and conventional Duvillaret algorithm. The comparison between the results is further analyzed for evaluating our method, the root-mean-square error (RMSE) between the two extraction methods and the standard value is calculated, and the extraction results of the transmission model, the Duvillaret algorithm and the standard value extracted from the parallel and smooth sample are compared at the same time. The comparison results show that, compared with the Duvilaret algorithm, the refractive index and absorption coefficient of the rough sample extracted by the transmission model method has smaller deviations from the standard value, which reduces the influence of roughness on the extraction result to a certain extent and has higher accuracy. Therefore, under the condition of known roughness and average thickness of the surface rough sample, this model can be used to more accurately extract the optical parameters of the substance in the terahertz time-domain spectroscopy system. The research results will greatly promote the practical application of terahertz optical parameter extraction technology.