Pulse or digital communications – Bandwidth reduction or expansion – Television or motion video signal
Reexamination Certificate
1999-12-21
2003-10-07
Kelley, Chris (Department: 2713)
Pulse or digital communications
Bandwidth reduction or expansion
Television or motion video signal
C345S543000, C345S547000, C345S544000, C345S555000
Reexamination Certificate
active
06631164
ABSTRACT:
FIELD OF THE INVENTION
The invention relates to an addressing process for the storage of data in memories of SDRAM type.
BACKGROUND OF THE INVENTION
Digital video processing typically requires that video images be stored in memory. Two types of memories are commonly used for storing video images:
Static Random Access Memories (SRAMs). The areas of these memories are directly accessible from the address bus.
Dynamic Random Access Memories (DRAMs), or, if they are synchronous, Synchronous Dynamic Random Access Memories (SDRAMs).
Dynamic memory is split into pages and access to a new page (corresponding to a change of page) requires an access time referred to as a “time penalty”. A memory address is composed of a first part referred to as the page address or “row address” and of a second part referred to as the binary address or “column address”. The terms in quotation signs mentioned here are the terms commonly used in technical sheets for DRAM or SDRAM type components.
Video processing applications, which are becoming increasingly common, need to access Image blocks, that are composed of a set of pixels from an image.
Such is the case, for example, for digital video data compression according to the MPEG standard, the acronym standing for Motion Picture Expert Group, where the elementary coding size of video data is an image block constituting a macroblock. During motion estimation, a step of the digital video compression process, a current image macroblock is compared with macroblocks of a reconstructed image that are stored in an SDRAM type memory to determine which reconstructed macroblock allows for the best correlation with the current image macroblock.
In the MPEG standard, the macroblocks are sets of data consisting of four blocks of 8×8 luminance data elements and, for the 4:2:0 compression format, a macroblock is represented as two blocks of 8×8 chrominance data elements, emanating from a section of 16×16 elements from the luminance component of the image.
Typically, motion estimation calculations are performed on the luminance values alone. The relevant blocks are then just the luminance blocks of overall size 16×16. These blocks are stored in the form of blocks of 16 pixels×8 lines, the luminance blocks of overall size 16×16 being previously separated into two blocks of 16 pixels×8 lines. These new blocks correspond to a first block of the upper frame and a second block of the lower frame in image mode grouping together two frames or to a first block grouping together the first 8 lines and a second block grouping together the next 8 lines of the same frame in frame mode. These luminance blocks of 16 pixels×8 lines are called, hereinafter, luminance semi-macroblocks or simply semi-macroblocks as to indicate that the grouping together of 2 luminance blocks of 8 pixels×8 lines, the term luminance macroblock being used to denote, in the MPEG standard, the 4 luminance blocks.
It is generally necessary to retrieve, from the memory, image blocks of a size greater than blocks of 16 pixels×8 lines. The size of the retrieved blocks is typically defined by the application using such blocks.
For example, the calculation of the motion vectors is carried out with a resolution of half a pixel requiring the processing of image blocks having an additional line and a column, that is to say of dimension 17 pixels×9 lines.
Another example relates to the search windows in a reconstructed image, in which the correlation of the blocks is performed having bigger or smaller dimensions depending on the process used for motion estimation. For a process designating, in a first step, a use of a search window on the basis of a motion vector, it is generally necessary, in a second step, to refine the search to perform a local adjustment, such a refinement being carried out on a window of small dimension. In one example, these search windows have a dimension of 24×12, corresponding to a horizontal excursion of ±4 and a vertical excursion of ±2 for blocks of 16×8 pixels.
The access time to an entire block of larger size is not generally optimized, requiring access to several pages of the SDRAM memory as to read or write the values of pixels constituting this block.
FIG. 1
represents luminance macroblocks of size 16×16, a first 16×16 luminance macroblock 1 and a last 16×16 luminance macroblock 2 of one and the same horizontal row, that is to say of a horizontal succession of macroblocks over an image width. In the general case, a row of macroblocks corresponds to a slice, as defined in the MPEG standard. Hereinafter, the term row will be particularly used to define a succession of blocks or of semi-macroblocks over an image width, the term slice being reserved for the macroblocks.
Indicated on the abscissa axis are the pixel numbers and indicated on the ordinate axis are the line numbers. 16 video lines over an image width of 720 pixels correspond to 45 macroblocks of 16 pixels (720:16).
The semi-macroblocks are stored one after another according to a television type scanning, as indicated hereinafter.
FIG. 2
represents memory pages 4, 5, 6, 8 such as they are successively addressed during the storage of the semi-macroblocks. As indicated previously, the 16×16 macroblocks are stored into two parts, the upper blocks and the lower blocks.
If the upper blocks correspond to one frame and the lower blocks correspond to the next frame, that is to say in image mode, the upper blocks are stored one after another in a memory space and the lower blocks one after another in another memory space, each memory space thus corresponding to a frame.
If the upper and lower blocks correspond to one and the same frame, that is to say in frame mode, they are stored one after another, that is to say firstly the upper blocks of a row (of semi-macroblocks) then the lower blocks corresponding to the next row. The next frame is stored in another memory space.
The storage process is described for the frame mode and can be generalized without difficulty to the image mode. It is applied to each of the memory spaces, taken separately, by considering, for each of them, only the semi-macroblocks which are stored therein.
A page 4 can store, in our example, 8 semi-macroblocks MB referenced
3
. The storage of the next 8 semi-macroblocks MB is performed in the next page 5 and so on and so forth. There is therefore, every 8 semi-macroblocks, a page change or skip
7
referred to as a “page miss” in the technical literature. Semi-macroblock No. 45, which corresponds to the end of the first row, is stored in the middle of a page referenced
6
, more exactly as 5
th
semi-macroblock out of the 8 semi-macroblocks of the page.
The next semi-macroblock, which corresponds to the semi-macroblock at the start of the second row of the frame, is stored at the start of a new page 8 in the memory, the start-of-page storage of a start-of-row semi-macroblock being the simplest solution for handling the addresses.
FIG. 3
shows the drawbacks of such a prior art. The memory space is represented schematically as a function of the size of the semi-macroblocks
13
stored. The emboldened lines
12
correspond to the boundaries of pages, pages of dimension 8 blocks of 16×8 pixels. Three overlaid complete pages are represented corresponding to groups of overlaid semi-macroblocks of three successive rows in the frame. Requiring to access image blocks of greater dimension than that of the semi-macroblocks stored, it is necessary to access several pages so as to read an image block. Thus, access to the block of 17 lines×9 pixels, referenced
9
, requires 4 page changes, access to block
10
of the same dimension, 2 page changes and access to block
11
of dimension 24 lines×12 pixels, 6 page changes. This calculation takes Into account the first page change that has to be performed, to access the block.
In this configuration, the maximum number of page changes for a block of dimension 17×9 is 4, the maximum number of p
Allard Jean-Marc
Plissonneau Frédéric
Sorin Alain
Fogelson Joel M.
Kelley Chris
Kurdyla Ronald H.
Parsons Charles
Thomson Licensing S.A.
LandOfFree
Addressing process for the storage of image blocks does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Addressing process for the storage of image blocks, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Addressing process for the storage of image blocks will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3120157