Due to the disadvantages of the total sky imager(TSI)

there are a large area of occlusions and a certain extent of distortion in the ground-based cloud image applied into monitoring the sky conditions and predicting solar radiation

which would result in the inaccuracy of cloud detection

cloud classification and solar radiation forecast based on ground-based cloud image. Therefore

a mirror gradients algorithm is proposed to restore the distribution of cloud in the image. The position of the sun in the image is firstly calculated to mark the occlusions automatically. Then the ground-based cloud image is filled and inpainted with mirror gradients algorithm based on the color feature of cloud. Next

the white pixels around the sun are classified into cloud or cloudless sky according to the attenuation of solar radiation

and the grey values of the over-exposure cloudless pixels are adjusted. The experiment results show that the proposed pre-processing method could restore the cloud distribution condition of the sky quickly and it is better than some other published methods used in photovoltaic system. It gives the strong support for the following research such as the change of climate and solar irradiance forecast.