Classifications

• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
  • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
  • H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
  • H04N19/124—Quantisation
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
  • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding

Definitions

• Samples of a signal transmitted by PULSE CODE MODULATION are usually coded within a whole absolute scale, e.g. as one of 256 values represented by one byte.
• the caused tremendous data stream may be reduced by preceding the extremes MIN, MAX of a block by two 8bit-codes.
• Each pel within the black is codable exactly between such precoded ("pre-dicted") ranges by a RANGE REDUCED PCM (RRPCM).
• Convergence is achieved by diminishing the overhead of those PRECODES to a NEAR ZERO VALUE: Like the well ordered elements of a segmented (broken) ceramic are reconstructable without any precodes, using the SURFACE INFORMATION only, it must be possible to reconstruct a comparable INFORMATION SOLID by its surface information, at least with few overhead only, being able to utilize CONVERGENTLY DIMINISHED RANGES, being often distributable upon multiple dimensions.
• a first experimental digital 8mm-consurmer Video was presented, by SONY Corp., Tokyo, compressing a 215-Mbps signal to a bandwith of 25Mbps "without visible degradation" , called ADAPTIVE DYNAMIC RANGE CODING /5/.
• ADRC utilizes a PRECODING of the minimum and the maximum to obtain reduced block ranges, allowing an exact, but RANGE REDUCED PCM.
• a range acaptively fins or coarse quantization diminishes that reduced bit rate again,while coding the picture and changed areas to a subsequent picture.
• the extremes MIN,MAX are precoded by two 8bit words, limiting the rentable block size of a plane to6x6pel.
• R1 Any limited area contains its extremes, the minimum (MIN) and the maximum (MAX), a common principle indicated by WEIERSTRASS (FIG. 1).
• R2 The smaller an area is, the smaller the range of a seals becomes between its 2 extremes. This was already indicated by CAUCHY for exactly continuous areas but may always be utilized as a statistical fact.
• R3 Ths smaller the scales, the less the average bit rate to code the areas separately by RANGE REDUCED PCM (RRPCM).
• FIG. 1 PA Segments, coded by RRPCM ars reconstructable like segments of a broken solid, especially utilizing ths SURFACE INFORMATION by offset minimization or even the knowledge based block specific offset code of the area. 2. Detaining those rules
• R1 to R3 must be detailed to be suitable for optimized SIGNAL SEGMENTATION.
• Ths range of an area is definable alternatively by the following parameter sets :
• MIN and R may be replaced by MAX and -R (inverted scale directionl). Knowing only one of these parameter sets, a RANGE REDUCED SCALE is defined to achieve an exact but RANGE REDUCED PCM (RRPCM-) coding of each signal area.
• RRPCM- RANGE REDUCED PCM
• FIG. 2 a According to version 1., ths most simple realization of defining a RANGE is the prseoding and transmission of both extremes MIN and MAX of a defined area (e.g. fixed block). Patent applications of SONY Corp. as mentioned in ref. /5/, are based upon this principle. The RANGE REDUCED or RANGE ADAPTIVE coding of ref./5a,b/ is called there ADAPTIVE "DYNAMIC" RANGE CODING (ADRC). But, since stored quantities of signal points are known "STATIC" SIGNAL QUANTITIES, this slight correction will help to utilize some aspects of STATIC FIELDS /1/ for the new method, to enable furthermore considerably reduced bit rates.
• RRPCM remains a relative coding for each block.
• Their SYNCHRONIZATION PROBLEM is solvable according to R4.
• 2.1. Basic reflections about the reduction of the overhead a) According to 2c) the (dynamic) range of a secarately coded block decreases continuously. Utilizing only their preceded range R(i), all blocks 5(i) are codable secarately point by pont.
• a successive proceeding may be used in every PARTIAL AREA, e.g. 32x32, (32x16,) 16x16, (16x8, )...2x2 pel, with successively reduced scales.
• PARTIAL AREA e.g. 32x32, (32x16,) 16x16, (16x8, )...2x2 pel, with successively reduced scales.
• MIN(h+1), MAX(h+1) and R(h+1) are codable within the (normally already reduced) scale
• Object of the method is to utilize a CONVERGENCE of the whols resulting bit rate b consisting of their components, utilizing hierarchical nestings. a) Using (MIN, R), the bit rate b (in bit/pel) consists of
• br1 BR1/BS computed in bit/block: 0.047 0.153 0.75 (absolute values) br2 by hierarchical method as measured: 0.032 0.13 0.55 (mean value)
• Each value t(i) may also define one side of a square!
• T shows ths delay fault as caused by the system, uhen transmitting an ideal edge (small FIG.3):
• the ideal edge is characterized by sharp CUMULATION POINTS HP1 , HP2.
• Transmitted or unideal edges degenerate to functional distributions within EM21 , EM22 along ths ordinate f(t).
• Such CUMULATION FUNCTIONS(DISTRIBUTIONS) characterize more or less each area(FIG.3). Their deneities and widths are measurable like T and HP along all coordinates. Especially they may be utilized for variable lenghts codings (HUFMANN CODES).
• FIG. 3 d Using such a SCALE or a fixed code length CL instead, a cumulated quantity of points is searched,where CL is valid for.
• CL is valid for.
• two run length codes (x,y) are selectable to define rectangular areas or simple squares.
• Block range precoding witn automatic relevance selection
• FIG. 5 shows a hierarchical plane division (QUAD TREE) as ussd by BRIT. TELECOM for its method RECURSIVE BINARY NESTING (RBN) /2/.
• the basic principle will be combined with R2 to result in a successive hierarchical division of the area, resulting now in sucessive code length reductions (for planes and 3D-spaces):
• FIG. 5 a One transmitted bit informs the receiver if a block, e.g. of 64x64 pel (or even a whole frame!), dhly needs a PCM-scale codable by less than 8 bits.
• an INTERRUPT CODE (1bit) informs the receiver of the result of an ANALYSIS, made by the transmitter, if the optimum is reached.
• CONTOUR SELECTION may be accomplished by a CONTOUR (better new: CORRELATION LINE-) PREDICTION, as given by ZSCHUNKE /4/, for BLOCK SYNCHRONIZATION according to 2.4c. Since such relevance chainsmay also be selected in the next picture of a sequence, the RANGE CORRELATION OF CL is advantageously measurable: Effects like flickering water or fade cuts, preventing movement detections, still have correlations in the luminance RANGE !
• a further dimension may still be divided in common (or separately along the z-axis) by such (1bit-)codes, depending on ths degree of (z-)changament within a 3D-block, containing several pictures. Since a sum of all elementary differences caused by noise is ZERO, the mean value along z may be coded to obtain static plane blocks, noiseless codes being distributable upon all blockwiss static pictures.
• Block synchronization by including one overlapping point by including one overlapping point (one difference!)
• the methods 2.4. are suitable, enabling a NEAR ZERO FAULT along the borders, correctable by a,now RANGE LIMlTED AND CUMULATION DEPENDENT HUFFMAN CODE. That kind of DIFFERENCE MINIMIZATION along block borders is comparable to the reconstruction of a ceramic, segmented by falling down. The smaller the blocks, the greater the average faults (differences to be coded).
• the first block is codable with its RRPCM-codes OFR within CL1.
• One common point P12 (e.g. the right upper corner of the precoded block 31) is chosen at the border of those neighbouring blocks.
• the distance P12 - MIN2 defines a REFERENCE WORD RW(12).
• the value RW may be coded as well using a positive 2-bit-word C2 within CL2 from the still unknown MIN2.
• C2 is known and MIN2 may be calculated as an exact PCM value within 62 (the address of MIN2 in 82 should also be coded! ) .
• All other points P2(j) of S2 are codable new as a pure positive offset to , MIN2. Since each point Pl(j) of B1 is chcosable as RW, it may even be MINI itself! e) MIN2 is computed as its PCM- word and all points of B2 as an offset, to MIN2.
• VECTOR RULES are applicable, constructing BRIDGES betueen RW(i) and MIN(i), especially between (small!) adjacent blocks. - see FIG. 5.
• the (highly redundant) range of a border of 1pel thickness was included in the measures, meaning the range R(4pel) of a 2x2 block measured for each 1x1 -pel-block instead of the correct difference between 2 pels only:
• Equivalent methods may use possible contradictions of an arbitrary sign and its real ranges R(h), R(h-1)..., to invert it, or even a statistical method.
• the VALUE of one extreme (normallyMIN, aternatively MAX) was precoded as a quasi sign-free difference from RW. Its VALUE would be coded redundantly twice as a RELATIVE ZERO, while the address is unknown. If the length CLA, used for the addressing of the selected extreme MIN or MAX is less than CL2, the extreme should advantageously be addressed, the complete code containing now the address and the difference D12. a) Tabls of the amount of bit according to 3.6., but coding now one address in each block for one EXTREME (MIN2, M ⁇ X2) instead of one MIN-code (relative block-zero!). Where the cods CLA for the address exceeds CL2, CL2 is taken (*).
• Range reduced HUFFMAN CODES may serve to diminish the bit rate furthermore, but even any other (nonexact) method. 4. Including other methods
• FIG. 6 b An additional knowledge of MAX as a regional (box-)vector, with block address and functional value, uill result in an additional address overhead, but offers advantages while the boxes are not too small, since CORRELATIONS are easily to be found in subsequent pictures, also using CUMULATIONS.
• the RANGE NUMBER is precodable by 0 or 1 (continuing with additional 0 or 1): One bit may precode, if the value is part of the first or the second (the residual) quantity, one excluding the other. The quantities are sharable again, if suitabls. Utilizing such EXCLUSION PRINCIPLES, the average bitrate per block may be reduced, e.g. coding 0..1 with 1bit and 2..9 with 3 bit (one added for the range number) instead of coding 0...9 constantly with 4bit.
• a scale R may consist of R2 + R3 even of R3 + R6 only, if no value lies within another RANGE NUMBER, which are all defining EMPTY RANGES!
• PROBABLE HUFFMAN- or RANGECODES(4.2) of short lenght are reserved for CUMULATIONS, longer words for less dense sections, even approximately quasi EMPTY RANGES R(i,j,k%) of a domain.
• Real objects contain sharp contours.
• An ideal contour selects two quantities, e.g. a quantity of white and a quantity of black points. In this case, there is no value existing between black and white! The whole range in between is an
• the nearest RIGHT or LEFT point may be supposed to be the true function value (A-prediction).
• A-prediction Each PREDICTION, made by a surrounding, evan DIRECTED to the edge, might prevent a kind of too abrupt jerkeyness.
• the transmitter is able to code a measured range of such an ERQ in the same way as ths mentioned real ranges when using the LOCAL GRADIENT!
• FIG. B As mentioned in 3.3d, the values of a plane containing addressss and function values may be regarded as VECTORS, having 3 coordinatss. This is suitabls for MIN(i) and MAX(i), while the other vectors, as a set of parallels, are codable subsequently by making address information redundant (degenerated vectors).
• All kind of DISPLACEMENT VECTORS may be measured normally as known from frame to frame, resulting in virtual CORRELATION LINES (FIG. 9).
• the blocks may now be estimated,their range being represented by an angle: often becoming smaller if the estimation is true (TRUE: the point has been found within) and wider, if not, until TRUE.
• TRUE the point has been found within
• a correlation point may be found again within subsequent pictures to compose a GREAT BOX OF SEVERAL SUBSEQUENT PICTURES (FIG.10), nestable in analogy to 3.1. Generally the highest deviation defines the size (CL!) of an initial "big box”. Smaller, CURVED AREAS are selectable like edges (FIG.5) or between correlation lines, as shown by the black arrows (boxes/areas) all treatable according to 3, since their components are quite normal signal functions, RANGE PRE-DICTIONS being applicable for every block-function,treating components like a luminance. Such a measured (or even statistically) pre-dicted RANGE (BOX) OF SUBSEQUENT PICTURES is mentioned in /5d/,prior /7b/. Now (new) the EVENT of FIG. 10 is also advantageously nestable according to 3, FIG. 5 (FIG. 2, 3, 4) and FIG. 9!
• the new method is able to reduce exacthess in subsequent coordinates (components) bit by bit:
• All kind of DISPLACEMENT VECTORS may be measured normally as known from frame to frame, resulting in virtual CORRELATION LINES (FIG. 9).
• the blocks may now be estimated,their range being represented by an angle: often becoming smaller if the estimation is true (TRUE: the point has been found within) and wider, if not, until TRUE.
• TRUE the point has been found within
• the address range of all components may be mininized.
• a correlation point may be found again within subsequent pictures to compose a GREAT BOX OF SEVERAL SUBSEQUENT PICTURES (FIG.10), nestable in analogy to 3.1. Generally the highest deviation defines the size (CL!) of an initial "big box”. Smaller, CURVED AREAS are selectable like edges (FIG.5) or between correlation lines, as shown by the black arrows (boxes/areas) all treatable according to 3, since their components are quits normal signal functions, RANGE PRE-DICTIONS being applicable for every block-function, treating components like a luminance. Such a measured (or even statistically) pre-dieted RANGE (BOX) OF SUBSEQUENT PICTURES is mentioned in /5d/,prior /7b/. Now (new) the EVENT of FIG. 10 is also advantageously nestable according to 3, FIG. 5 (FIG. 2, 3, 4) and FIG. 9!
• the new method is able to reduce exacthess in subsequent coordinates (components) bit by bit:
• HIGH RELEVANCE may be given by a SET OF ABRUPT DIFFERENCES (offsets OFR), selecting contours of natural areas for an ARTIFICIAL VIEWING as well as for future RANGE PREDICTION by defining a FUNCTION CODE LENGTH CLF as well as an ADDRESS CODE LENGTH CLA(x.y) for "displacements", defining BOXES.
• the primary boxes may be devided in all (three spatial and fuction) axes (components), using (together or separately) the rules R1 to R3 (FIG. 6).
• PREDICTED CODE LENGTHS CL(x,y,z; i,j,k) may be taken es FIRST REFERENCES for DISPLACEMENTS, being correctable bottom-up as well as precisable top-down, defining (hierarchically nested) BOXES to contain all possible values!
• c) If a picture fades in or out, already relatively extreme valuss of CL give a good reference for a defined object, especially combined with CUMULATIONS, as mentioned above, while the conventionally defined correlation may even indicate maximal z-difference (meaning maximal disporrslationl) at the same place, e.g. when detecting a light which fades out within a dark plane. While imagss of flickering reflecting waves or fire prevent to obtain results by direct correlation measurement, the CL-CORRELATION method will still succeed.
• ADRC Adaptive Dynamic Range Coding Scheme
• ADCR Range reducing by MIN and MAX: 2, line 1S ff. by also
• 3D-ADRC ranges incl.time axis: SUMMARY of the invention, p.1ff. - claim 1
• Too-down reduced ranges and displacements e.g. p.5
• FIGURE 1 Range of a planar area between its extremes MINIMUM and MAXIMUM.
• FIGURE 2 Typical linear luminance function of an edge: A simple linear
• FIGURE 5 Hierarchical plane division of an area, by BRITISH TELECOM (modified).
• the CONTINUATION PRINCIPLE is utilizing punctual values on the borders.
• FIGURE 5 Block with positions of MIN and MAX in the 3D-FUNCTION+ADDRESS SPACE.
• FIGURE 7 Within monotonuously (not) rising or falling function values or address spaces or at an EDGE, a coded space (function part) may be excluded. The remaining address spacs is continuously diminishing.
• FIGURE B Different kinds of volumes, the first one showing the extremes
• FIGURE 9 Following an object using its scales defined by CL and/or DISTRIBUTIONS of its CUMULATION POINTS within bottom-up and top-down varying angles (planes): If a statistically predicted range (box, angle) is FALSE, the angle (box, range) is widened, otherwise hierarchically diminished.
• FIGURE 10 Image sequence with static and dynamic areas (white and dark arrows).

Landscapes

• Engineering & Computer Science (AREA)
• Multimedia (AREA)
• Signal Processing (AREA)
• Compression, Expansion, Code Conversion, And Decoders (AREA)
• Compression Or Coding Systems Of Tv Signals (AREA)
• Image Processing (AREA)

Applications Claiming Priority (6)

Publications (1)

Family

ID=27222739

Family Applications (1)

Country Status (2)

Families Citing this family (1)

Family Cites Families (7)

• 1989
  • 1989-08-23 WO PCT/EP1989/000990 patent/WO1990002465A2/en not_active Application Discontinuation
  • 1989-08-23 EP EP19900900125 patent/EP0432222A1/de not_active Withdrawn

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events