High-quality ghost imaging （GI） at low sampling rate is of great importance for scientific research and practical applications. Therefore， the reconstruction of high-quality images under low sampling rate conditions remains the focus of GI research. In this paper， a high-quality passive compressive ghost image reconstruction algorithm was proposed， called PCGI-LRC. Based on the assumption that the matrices stacked with nonlocal similar blocks of an image have low-rank and sparse singular values， a joint iterative solution of the least squares problem was demonstrated theoretically and experimentally and the low-rank approximation problem of the nonlocal similar blocks can achieve high-quality ghost images under low sampling rate conditions （6.25%-50%）. Moreover， the experimental results show that the proposed algorithm outperforms the GI based on sparse basis constraints （GI-SBC） and GI based on full variational constraints （GI-TVC） algorithms regarding peak signal-to-noise ratio （PSNR）， structural similarity coefficient （SSIM）， and visual observation. Information of the target is preserved while the reconstruction noise is suppressed； the PSNR is improved by more than 1.1 dB and the SSIM improvement is higher than 0.04 dB.