CLOUD DETECTION BY INTER-BAND PARALLAX AND A-CONTRARIO VALIDATION

The presence of clouds often causes detection errors in automatic optical satellite images analysis. Cloud detection is thus an important pre-processing step. This work exploits the apparent movement of clouds relative to the ground that is observed between bands due to the parallax effect in push-broom satellites. A simple optical flow is computed by correlation of the gradient direction on a local window. Then, a statistical validation test based on the a-contrario theory is used to reject spurious detections. The proposed method achieves performances in the same order of magnitude as the state of the art, while relying on less input information.