Aiming at the process of low-dose photon counting imaging with Poisson-Gaussian mixed noise, a sparse reconstruction method of integrating data fidelity term and sparse constrait term is proposed. Firstly, based on the hypothesis that Poisson and Gaussian noise are mutually independent, the sparse reconstructing objective function based on integrating data fidelity term and sparsity constraint term is established. Based on patch clustering, the improved greedy algorithm is applied to implement sparse decomposition and dictionary update. Finally, a clean image is obtained by alternating iteration. Contrast experiments on images corrupted with strong Poisson-Gaussian mixed noise show that the average PSNR of image reconstructed by the proposed method increased by 5.5% more than those of the contrast methods, moreover, their MSSIM increased significantly. The experiment results demonstrate that the proposed method has better image restoration and denoising effect for low photon counting image with strong Poisson-Gaussian mixed noise.