The ability to replace missing data in images and video is of key importance to many application fields. The generalpurpose algorithm presented here is inspired by texture synthesis techniques but is suited to any complex natural scene and not restricted to stationary patterns. It has the property to be adapted to both still images and image sequences and to incorporate temporal information when available while preserving the simplicity of the algorithm. This method gives very good results in various situations without user intervention. The resulting computational cost is relatively low and corrections are usually produced within seconds. KeywordsImage and video processing, restoration, constrained synthesis, non-stationary and non-parametric Markovian models

No abstract