Large number of sampling and low signal-to-noise ratio (SNR) limit the practical application of ghost imaging process. The traditional ghost imaging using Gaussian random field imaging ，however, speckle noise is introduced by the non-orthogonality of Gaussian matrix , which decreases the SNR and efficiency. Therefore, local Hadamard matrix is used to measure the target, and the clear image of low-resolution image is utilized as a prior knowledge for high resolution image to achieve denoising, which can effectively eliminate the correlated noise. Simulation and experimental results show that, comparing with the traditional reconstruction methods, the proposed method can increase the SNR and imaging efficiency.