Method for motion estimation and bandwidth reduction in memory and device for performing the same
View Patent ↗A device and a method are disclosed for motion estimation and bandwidth reduction in a memory. The device includes a memory for storing a plurality of frame data, a controller connected to the memory, a first motion estimation processor for performing a coarse-tuning operation and a second motion estimation processor for performing a fine-tuning operation. Similarity between a reference frame and a current frame is calculated based on the averages of every two adjacent pixels in the reference macroblocks and current macroblocks. The amount of calculations for determining motion estimation is greatly reduced, and bandwidth in utilizing the memory is accordingly reduced as motion estimation is advantageously achieved.
1. A motion estimation method comprising the steps of:
(a) displaying an image in a plurality of frames corresponding to a plurality of time periods; wherein each frame further comprises a given number of pixels, each pixel represented by two-dimensional abscissa and ordinate coordinates;
wherein each frame further comprises at least one macroblock having a lesser number of pixels than said given number of pixels of said frame;
wherein a current frame is one of said frames in a current time period of said plurality of time periods;
wherein said current frame further comprises at least one current macroblock having a lesser number of pixels than said given number of pixels of said current frame;
wherein a reference frame is one of said frames in a time period of said plurality of time periods prior to said current time period for said current frame;
wherein said reference frame further comprises at least one reference macroblock having a lesser number of pixels than said given number of pixels of said reference frame;
(b) determining calculated pixel units of said current macroblock, wherein one calculated pixel unit is an average of two adjacent pixels;
(c) repeating step (b) for all pixels of said current macroblock;
(d) determining calculated pixel units of one macroblock of said reference macroblocks;
(e) repeating step (d) for all pixels of said one reference macroblock;
(f) multiplying each calculated pixel unit of said current macroblock by a corresponding calculated pixel unit of said one reference macroblock resulting in a plurality of multiplying values;
(h) summing the plurality of multiplying values resulting in a sum of correlations (SC);
(i) shifting abscissa and ordinate coordinates of said one reference macroblock by corresponding abscissa and ordinate shift values resulting in a shifted reference macroblock;
(j) determining calculated pixel units of said shifted reference macroblock;
(k) repeating step (j) for all pixels of said shifted reference macroblock;
(l) multiplying each calculated pixel unit of said current macroblock by a corresponding calculated pixel unit of said shifted reference macroblock resulting in an additional plurality of multiplying values;
(n) summing said additional plurality of multiplying values resulting in an additional SC;
(o) repeating steps (i), (j), (k), (l), (m) and (n) for each reference macroblock resulting in a plurality of additional SCs; and
(p) selecting a selected SC from a group comprising the SC and said additional SCs, wherein the selected SC has a larger value than another of the group.
2. The method of claim 1 wherein said reference frame and said current frame are formed by even lines and odd lines.
3. The method of claim 2 further comprising determining a top field motion estimation using pixels in said even lines.
4. The method of claim 2 further comprising determining a bottom field motion estimation using pixels in said odd lines.