Fast multi-frame motion estimation with adaptive search strategies

A method comprising:

• performing first motion estimation with respect to a current frame of a sequence of frames of image data, the first motion estimation being performed with reference to a first set of reference frames, said first set of reference frames preceding said current frame in said sequence of frames of image data;
• evaluating a result of said first motion estimation;
• and
• determining, based on an outcome of said evaluating, whether to perform second motion estimation with respect to said current frame, said second motion estimation to be performed with reference to a second set of reference frames different from said first set of reference frames, said second set of reference frames preceding said first set of reference frames in said sequence of frames of image data,
• wherein said first motion estimation includes:
• determining a plurality of candidate motion vectors for a first pixel block in said current frame, the candidate motion vectors including at least two of the following:
• (a) a motion vector for a pixel block that adjoins the first pixel block;
• (b) a motion vector for a pixel block that is a superset of the first pixel block;
• (c) a motion vector for a second pixel block that is part of a reference frame that is a last frame in said first set of reference frames, said second pixel block having a position in said reference frame that corresponds to a position of the first pixel block in said current frame;
• and
• (d) a scaled motion vector for a third pixel block that is part of another reference frame, said another reference frame preceding said last frame in said first set of reference frames, said third pixel block having a position in said another reference frame that corresponds to a position of the first pixel block in said current frame;
• determining a number of said plurality of candidate motion vectors that are the same;
• and
• comparing said number to a threshold;
• and wherein said first motion estimation further includes:
• determining a characteristic of a frame of image data that is prior to said current frame in the sequence of frames;
• and
• selecting, on the basis of the determined characteristic, a block-matching search pattern for use with respect to said current frame, said prior frame immediately preceding said current frame in said sequence of frames, wherein said determining said characteristic of said prior frame includes analyzing a statistical distribution of motion vectors for said prior frame, wherein said selecting includes selecting between a hexagonal search pattern and a cross search pattern said first set of reference frames includes three reference frames that immediately precede said current frame in said sequence of frames of image data and wherein said second set of reference frames includes two frames that immediately precede said first set of reference frames in said sequence of frames of image data.