Text recognition during video compression
View Patent ↗A method for encoding an image of a video stream divided into macroblocks comprising determining a maximum of variance of luma blocks in a macroblock, determining if the maximum of variance exceeds a predetermined threshold, determining if the motion vectors of the macroblock are in correlation with a reference macroblock, determining if the macroblock is not an edge macroblock and setting a first quantization level for the macroblock if it meets these conditions and setting a second quantization level for the macroblock if it does not meet these conditions, where the second quantization level is greater than the first quantization level.
1. A method to encode an image divided into macroblocks, each macroblock including a plurality of luma blocks and a plurality of chroma blocks, wherein each luma and chroma block includes an array of pixels, comprising:
determining a maximum variance of luma blocks in a macroblock;
determining if said macroblock is in correlation with at least one reference macroblock by computing a correlation between motion vectors of said macroblock and motion vectors of said at least one reference macroblock;
quantizing said macroblock using a quantization level based on said maximum variance of said luma blocks and said correlation between said macroblock and said at least one reference macroblock, wherein said quantizing step includes setting a first quantization level for said macroblock if said maximum variance of said luma blocks is greater than a predetermined threshold and said macroblock is in correlation with at least one reference macroblock, and if each of a plurality of reference macroblocks has a maximum variance of its respective luma blocks greater than a predetermined threshold and is in correlation with its at least one respective reference macroblock, otherwise setting a second quantization level for said macroblock, wherein said second quantization level is greater than said first quantization level; and
repeating said determining and quantizing steps for each macroblock in said image.
2. The method of claim 1 , wherein a macroblock is in correlation with an reference macroblock if a horizontal motion vector of said macroblock is in correlation up to a predetermined threshold with a horizontal motion vector of said reference macroblock and if a vertical motion vector of said macroblock is zero.
3. The method of claim 1 , wherein said image is one of a P-picture or a B-picture of an mpeg video stream.
4. The method of claim 1 , wherein a reference macroblock is a macroblock prior to said macroblock and a number of reference macroblocks is at least three.
5. A method to encode an divided into macroblocks, each macroblock including a plurality of luma blocks and a plurality of chroma blocks wherein each luma and chroma block includes an array of pixels, comprising:
determining a maximum variance of luma blocks in a macroblock;
determining if said macroblock is in correlation with at least one reference macroblock by computing a correlation between motion vectors of said macroblock and motion vectors of said at least one reference macroblock;
quantizing said macroblock using a quantization level based on said maximum variance of said luma blocks and said correlation between said macroblock and said at least one reference macroblock, wherein said quantizing step includes setting a first quantization level for said macroblock if each of a plurality of reference macroblocks has a maximum variance of its respective luma blocks greater than a predetermined threshold and is in correlation with its at least one respective reference macroblock, otherwise setting a second quantization level for said macroblock wherein said second quantization level is greater than said first quantization level; and
repeating the determining and quantizing steps for each macroblock in said image.
6. The method of claim 5 , wherein said reference macroblocks are macroblocks prior to said macroblock and a number of reference macroblocks is at least three.
7. The method of claim 5 , wherein a macroblock is in correlation with an reference macroblock if a horizontal motion vector of said macroblock is in correlation up to a predetermined threshold with a horizontal motion vector of said reference macroblock and if a vertical motion vector of said macroblock is zero.
8. A method to encode an image divided into macroblocks, each macroblock including a plurality of luma blocks and a plurality of chroma blocks, wherein each luma and chroma block includes an array of pixels, comprising:
determining a maximum variance of luma blocks in a macroblock;
determining if said macroblock is in correlation with at least one reference macroblock by computing a correlation between motion vectors of said macroblock and motion vectors of said at least one reference macroblock;
quantizing said macroblock using a quantization level based on said maximum variance of said luma blocks and said correlation between said macroblock and said at least one reference macroblock, wherein said quantizing step includes setting a first quantization level for said macroblock if said maximum variance of said luma blocks is greater than a predetermined threshold and said macroblock is in correlation with at least one reference macroblock and if each of a plurality of reference macroblocks has a maximum variance of its respective luma blocks greater than a predetermined threshold and is in correlation with its at least one respective reference macroblock and if said macroblock is not an edge macroblock, otherwise setting a second quantization level for said macroblock wherein said second quantization level is greater than said first quantization level; and
repeating said determining and quantizing steps for each macroblock in said image.
9. The method of claim 8 , wherein a reference macroblock is a macroblock prior to said macroblock and a number of reference macroblocks is at least three.
10. The method of claim 8 , wherein a macroblock is in correlation with an reference macroblock if a horizontal motion vector of said macroblock is in correlation up to a predetermined threshold with a horizontal motion vector of said reference macroblock and if a vertical motion vector of said macroblock is zero.
11. A method to, encode an image divided into macroblocks, each macroblock including plurality of luma blocks and a plurality of chroma blocks, wherein each luma and chroma block includes, an array of pixels, comprising:
determining a maximum variance of luma blocks in a macroblock;
determining if said macroblock is in correlation with at least one reference macroblock by computing a correlation between, motion vectors of said macroblock and motion vectors of said at least one reference macroblock;
quantizing said macroblock using a quantization level based on said maximum variance of said luma blocks and said correlation between said macroblock and said at least one, reference macroblock, wherein said quantizing step includes setting a first quantization level for said macroblock if each of a plurality of reference macroblocks has a maximum variance of its respective luma blocks greater than a predetermined threshold and is in correlation with its at least one respective reference macroblock and if said macroblock is not an edge macroblock, otherwise setting a second quantization level for said macroblock wherein said second quantization level is greater than said first quantization level; and
repeating said determining and quantizing steps for each macroblock in said image.
12. The method of claim 11 , wherein a reference macroblock is a macroblock prior to said macroblock and a number of reference macroblocks is at least three.
13. The method of claim 11 , wherein a macroblock is in correlation with an reference macroblock if a horizontal motion vector of said macroblock is in correlation up to a predetermined threshold with a horizontal motion vector of said reference macroblock and if a vertical motion vector of said macroblock is zero.