Sampling number and detection signal-to-noise ratio (SNR) are two major factors influencing imaging quality. Combining the image's sparsity in the representation basis with the ghost imaging (GI) approach, GI via sparsity constraints (GISC) can nonlocally image the object even when the measurement number is far fewer than the Nyquist criteria required for the conventional GI reconstruction algorithm. The influence of receiving the system's numerical aperture and detection SNR in the test path to GISC is studied through experiments. It is also shown that the quality of GISC depends on the object's sparse representation basis.