The search for Earth-like exoplanets requires high-contrast and high-angular resolution instruments, which designs can be very complex: they need an adaptive optics system to compensate for the effect of the atmospheric turbulence on image quality and a coronagraph to reduce the starlight and enable the companion imaging. During the instrument design phase and the error budget process, studies of performance as a function of optical errors are needed and require multiple end-to-end numerical simulations of wavefront errors through the optical system. In particular, the detailed analysis of long-exposure images enables to evaluate the image quality (photon noise level, impact of optical aberrations and of adaptive optics residuals, etc.). Nowadays simulating one long but finite exposure image means drawing several thousands of random frozen phase screens, simulating the image associated with each of them after propagation through the imaging instrument, and averaging all the images. Such a process is time consuming, demands a great deal of computer resources, and limits the number of parametric optimization. We propose an alternative and innovative method to directly express the statistics of ground-based images for long but finite exposure times. It is based on an analytical model, which only requires the statistical properties of the atmospheric turbulence. Such a method can be applied to optimize the design of future instruments such as SPHERE+ (VLT) or the planetary camera and spectrograph (PCS - ELT) or any ground-based instrument.