Based on ghost imaging (ghost imageing, GI) and compressed sensing (compressed sensing, CS) theory, the influence of CS reconstruction algorithm on imaging performance of GI is investigated. Discrete wavelet transform is used as image sparse matrix, and the thermal light source intensity distribution with Gauss profile is used as measurement matrix. Quality of compressed GI is analyzed with total variation minimization algorithm based on augmented Lagrangian and alternating direction method (TVAL3), orthogonal matching pursuit (OMP), compressive sampling matching pursuit (CoSaMP) algorithm, gradient projection algorithm (GPSR_ basic). Mean square error, peak signal to noise ratio, matching degree and structural similarity index are taken as the objective evaluation criteria for image quality, and the reconstruction results of GI using four kinds of reconstruction algorithms are compared. Results show that when the compression ratio is 0.5, the reduction degree of TVAL3 algorithm is the highest, and the distortion of CoSaMP algorithm is the most serious. The reconstruction performance of GPSR_Basic algorithm is better than that of OMP algorithm.