Abstract: Locations of reference blocks of pixels of decoded interlaced video reference fields decoded from a compressed stream of video data are computed using motion vectors decoded from the same stream of video data. The reference block computed locations are de-interlaced with current blocks from current video fields decoded from the compressed streams of video data by low pass filtering light intensities of adjacent pixels of the reference and current blocks using low pass filter with coefficients adjustable according to discrete cosine transform (DCT) high frequency coefficients decoded from the coded stream. In addition, moving edges or objects of the de-interlaced blocks of pixels can be identified by comparing the decoded motion vector of the reference blocks to a threshold value. Then proper or more appealing light intensities for the moving edge pixels can be interpolated.

Abstract: A device including a two-dimensional convolution unit to perform spatial image filtering. A reference frame mirroring unit is connected to the two-dimensional convolution unit. A mean square error (MSE) decision unit is connected to the two-dimensional convolution unit to perform motion estimation by selecting the displacement that minimizes MSE.

Abstract: One embodiment includes a method that includes receiving a compressed video stream. The method also includes decoding a number of blocks of the compressed video stream to output a number of blocks of decoded video data. The decoding is based on at least one motion compensation vector. The method also includes deinterlacing the number of blocks of the decoded video data to output deinterlaced video data. The deinterlacing of one of the blocks of the number of blocks is based on the at least one motion compensation vector if a prediction error energy for the at least one motion compensation vector for the block is less than a threshold.