Currently, the core techniques of most video coders are based on blocked DCT (discreted cosine transform) transform. At very low bit-rates, video coders of this class always produce visual-sensible block artifacts. Recently, the video coders based on matching pursuit have been proved to have super coding efficiency, compared with the H.263 standard. However, the computation complexity is much too higher, because of laboriously searching optimal atoms in a redundant dictionary. A video coding algorithm based on undecimated wavelet dictionary is proposed, which incorporates multiresolution characteristics of the wavelet transform and assorts to filter the scheme between dictionary functions, to reduce the computation complexity of the original algorithm greatly. Considering the continuity of motion information in the neighboring frames, an entropy coding method of atom position based on a hygrid structure is given. The experimental results show that this algorithm keeps the original coding results and has the desirable application value.