GB2296401A - Motion vector encoder using spatial majority filter - Google Patents

Abstract

The interleaved sectional combination of an intra-frame spatial majority filter transfer upon scan line interpixel changes is matricised by a fully-compensated interframe change filter in the 'median' temporal-domain to result in a highly effective vector encoder giving exceptional transmission and display resilience to communication noise intrusion in video system distribution, and pixel bit rate or data rate reduction techniques. Purposes include digital video broadcasting and/or HDTV where an attribute change or motion class vector for up-conversion interpolation is derived on down-conversion at picture source for distribution to the destination receiver. The compromise reduction in spatial frequency for pixel vector representation of a scan line is enhanced by the removal of single pixel noise volumes to restore the source encoded signal. The spatial majority vector encoder also has instrumental purpose within video measurement, control, and display systems where the single frame ( DELTA T) delay is not significant. <IMAGE>

Description

DESCRIPTION IMPROVED 'MAJORITY' FILTER The majority vector encoder matrix (VEIl) of invention is for signal noise and error rejection on motion vectors or flags in digital television, video or vision systems, for any purpose of modulation envelope noise elimination by the neural vector logic-filter processing of binary vector or flag representations, for change or motion events with discrete (6Pai, AT) periodic reproducibility from pixel interframe change values in a linear predictive codification of monotonic vectors (pi,si) from adjacent kinematic frames/fields, for an interpolation or motion (N / m) vector, m-descriptor Ex, y,m), or 'one-bit' per-pixel-point image change detection, by scanning interval sampling and codification using either or both the temporal and/or spatial mode domain from one or more frame/field image sources (eg Y,Cb,Cr).

The majority vector encoder matrix for digital television codification is for an intermediate processing scheme network insertion into video bandwidth or bit-rate reduction for frame-rate limiting down-conversion or period doubling, motion vector channel and codec up-conversion systems, including motion-adaptive segment interpoiation. Location between the picture image source and the destination receiver-set display aleviates noise or ergodic intrusion by an adjacent codification matching of the change or motion vectors by interpixel and quaternary correlations.The finite impulse response (FIR) transversal or temporal-domain (st1) 'doublet ' section contains an intermediatary filter for improved majority motion or class vector (! / m) codifications of the interconnective Fourier quantum variable ( qi ') early in the temporal-domain (6tis) memory in transfer to quantum ('pus') on dilation.

The aim of continuous predictability is to eliminate odd singularity pixels, and therefore uncoded representations and/or generation error, noise or interference values by the spatial mode majority encoding of a finite temporal-domain impulse response filter (DIRF) section, in order to more precisely resolve with higher accuracy codifications, the maximum statistical likelihood values of image pixel change or motion for a binary class (m) vector or b(m) flag outcome from interframe subtractive or differential (PflD-value) data, discretely derived from successive kinematic video frames at or near the signal source.The canonical class formation from an intrafield quantum ('q'-- > 'p') dilation in the 'one-bit' temporal domain (6tl) is for two change or motion attributes to divide the interframe (N / m) vector inputs into canon classes tl-2] for the quad or quaternary [bl-2 or bl-4] partial lattice of the Median and majority filter outputs.

The change or motion class vector (m-descriptor) or b(m) flag value is a discrete image binary feed into the finite impulse response (FIR) transversal section for initial single quantum ('ql') values from prior sensory interframe vector (M / m) detection, as derived from a FIR type filter on the pixel frame input of digital television, video or vision systems.

Technological adaption for HDTV (high definition television), NPEG (motion picture expert group)/ISO, digital video broadcast (DVB) systems, etc, for the forward ('1') and back ('0') predictive frame interpolation ('0'/'1') in a destination system is for class motion-vector frame (k'n) reconstruction, incorporating the colour triad Median filter.

For specific video bandwidth or bit rate reduction, or compression systems a parallel picture signal (2\ T) frame-store, as a sequential frame interval delay of bi-stable (b-s) 'cache' pixel memory is used, in a multi-bit frame (n-1)/n-bit digital dynamic lane ( T) delay for the digital picture data from colour triad (eg Y,Cb,Cr) components, while the single binary change or motion vector (M / m) is automatically diverted into the vector encoder by-pass for the correlation matrices of the dual canon class (1-2] vector filters.

Previous methods are the earlier United States Patent number 4,651,207 of Bergmann et al, IBA Patent number WO 85/04542 Al, and by Jones supra, Patent Qpplications GB 2 265 783 (A) Bandwidth reduction employing a class vector channel, GB 2 265 784 (n) Video image motion classification and display, and GB 2 274 371 (A) Measurement and control using motion classification flags.

The application of a vector encoder matrix, including the intra-frame spatial mode of the majority filter, with adoptive correction eigenvectors'(ms), results in a regenerative flag enhancement, with canon classes (1-4], in video frame (k'n) interpolative reconstructions to source code the hex class 6 vector for Fallback b(6) flags, giving later motion adaptive segmentations (1-4] in the channel codec, and frame-rate up-convertion for picture display.

In accordance with a first aspect of the present invention, there is a method of achieving a majority vector encoder matrix using an intraframe spatial pixel recurrence filter having coded vector or flag correction, matricised within a temporal (T-D) domain (6tl) FIR transversal and multiple correlation sum function, with optional (m) secondary (6t2) feedforward in the (n,m) two-canal temporal (TT) transpose (n.m) equalisation, by canon class 2 vector (n), and flag b(2), in synchronous check-code compensation for the canon class 6 vector of the Fallback b(6) flag, to produce pixel noise volume reduction, for change or motion vector flag (b(m)) logic, for an m-descriptor, with parallel frame period (t T) delay, within television video compression systems for downconversion or up-conversion, in video distribution with frame delay compensation, for digital television, or for video, vision, or image processing from one or more Median signal sources.

Preferably, a two stage spatial filter with an intraframe vel-correlation capture filter is used to scan code the instigation of an adoptive vector velselection filter, direct and auto-correction (mO-l) output vectors (my), for flags (p,q), or singleton flags (b(m)) are produced, giving discrete singularity noise elimination and resilience for interframe vectors within television bit-rate reduction, video or image control or measurement systems, or to remove the vector noise volume of single pixel occurrences.

Preferably, for an image frame signal, a reduction in the occurrence is achieved in either or both of the intraframe spatial pixel to pixel domain, and/or the interframe temporal-domain frame to frame, for quaternary class (1-4) detection, with optional secondary temporal (6t2) equalisation for double sampling compensation on pixel-point motion vectors on the change (6Pa1, etc.) duration.

Still more preferably pixel majority vector encoder matrix is achieved using a finite (temporaldomain) impulse response or quaternary majority filter, with secondary temporal equalisation (n,m) in conservative graceful (dev n) compensation in frame allocation rebate by synchronising the two-canal or temporal transpose (TT) vectors (n,m) for Fallback b(6) flags of canon class 6 vectors, on video attribute change or motion, resulting in an elimination of singular vector or pixel flag occurrences in the temporal frame to frame and/or the spatial pixel to pixel vector, or in flag noise.

In a further preferred embodiment, a relative elimination of noise volume occurrences upon single pixel vectors or scan flags of an image frame sign, in both the majority spatial domain, and in the input finite temporal (T-D) domain (6tl) filter is achieved without compensation using a 'fully motive' singleton flag of Luma difference on the first matrix double correlation sum of the canon class 1 vector as a singleton flag b(l) on the filter output for machinevision, control and instrumentation.

In still a further preferred embodiment, there is a singleton class b(5) vector from canon class 1 motion conditions available in multiplex diversion from a plurality of secondary temporal (TT) transpose memory (6t2) delays in by-pass synchronisation by a diversion with the main (dot) frame (b-s) delay.

Preferably, an intraf rams Median hierarchial filter with multiple tree-section inputs for optional triad signal inputs is used.

In yet another preferred embodiment, for digital image or video frame signal an m-descriptor identifier matrix is used.

In still another preferred embodiment, for a digital image or video frame signal, a decoder mdescriptor matrix is used.

Preferably a motion vector (M/m) channel is used for transfer in down-conversion or up-conversion in bandwidth reduction or bit rate compression.

An error noise function (ENF) indicator may be used to give an indication of encoding effectiveness.

In accordance with a second aspect of the present invention, there is provided a majority vector encoder matrix with an intraframe spatial pixel recurrence filter having coded vector or flag correction, matricised with a temporal (T-D) domain (6t1) FIR transversal and multiple correlation sum function, with optional (m) secondary (6t2) feed-forward in the (n,m) two-canal temporal (TT) transpose (n,m) equalisation, by the canon class 2 vector (n), and flag b(2), in synchronous check-code compensation for the canon class 6 vector of the Fallback b (6) flag, producing pixel noise volume reduction, for change or motion vector flag (b(m)) logic, for an m-descriptor, with parallel frame period (AT) delay, within television video compression systems for downconversion or up-conversion, in video distribution with frame delay compensation, for digital television, or for video, vision, or image processing from one or more Median signal sources.

In accordance with a third aspect of the present invention, there is provided a device for implementing, within the intra-frame spatial scan, an auto-corrective majority filter with an instigation (mO) logic window for full replacement (m5) vector output in the one-bit (q) temporal-domain (T-D) or discrete (M/m) transfer.

Specific embodiments of the present invention will now be described1 by way of example only, with reference to the accompanying Figures, in which: Fig. 1 is an overview diagram of the duo neural channel networks to show the one-bit per triad signal processes for class vectors, in both the intra-frame spatial (6Pa1/2 / 6Pb1/2) mode and interframe FIR transversal (6t1) and a later temporal transpose (6t2) group (n,m), to compose the two canon classes 1 and 2 for the total singleton and the hex class 6 Fallback vector of the b(6) flag.

Figure 2. is a diagram of the intra-frame majority filter with one spatial (spa) ) interpixel memory, also showing an extension to the double (Pai.SPa) memory with single or twin sampling correction situated within the major temporal-domain (#t1) memory transfer delay of the FIR transversal.

Figure 3. shows the Boolean image property of a 4-frame transition diagram for the data communications network of the quad or quaternary canon classes C1--43, with the quin class 5 vector of canon class 2 vector synchronisation for the hex class 6 Fallback vector.

Figure 4. is the image-code truth table for the vector encoder matrix for classifications [1-6] upon a frame motion sequence.

Al-o shown is the outline schematic of the spatial majority and temporal compensation filters on an encoder matrix diversion by-pass from the main bi-stable (b-s) 'cache' frame (t T) memory delay.

FigureS 5 a/b are circuit diagrams of neural network functions and node interconnections for the intra-frame (#Pa1/2,#Pb1/2) spatial (ms) corrective majority filter within three frames (Pk(n), Pk(n-l), Pk(n-2)) of the temporal delays (#t1 & #t2) for the correlation compensation networks of the vector encoder matrix, and the (Figure 5 b) Median hierarchial tree.

Figure 6 a.b / c.d is a diagrammatic frame serial of temporal ETT] transpose vertical cross-sections from the 4-frame data cascade with classes Clto6] for motion adaptive segments Clto4 from video frames Pkn, Pkn-1, and Pkn-2.

Figure 7. are performance illustrations of the multiple correlation sum for the canon class 1 vector b(l) flags of the Luma difference occurrences for function comparison with the dual correlation NOT/AND partial sum for canon class 2 vector b(2) flags of Neana difference occurrences.

Figure 8. is a Mealy and Moore state diagramming on the machine program code for the 2stagesPatial majority of the logic window filter with Z-pole adoptive (mo-L) correction.

Figure 9. Error Noise Function indication of the (ENF) flag.

Figure 10 shows a decoder m,descriptor identifier matrix for a digital image or video frame signal.

The majority vector encoder matrix scheme (Figures 1, 2 5 and 8) is the synchronous neural combination of each of the temporal (#t1/2) and the improved spatial (#Pb1) filters of principally two separate image vector transfer functions. Each vector flag filter section has 2-pole dimensionality upon the interframe vector (M / m) from input detection [#:#q = > m, for 'q2']. The input FIR filter by a (M / m) transversal (it) of the temporal-domain (T-D) memory ('q1'-- > 'p1') delay transfers the neural transitional bi-codification of the Hamiltonian viterbi-code tree circuit (see Figures 1 and 5) to pair form the first order ('q' R 'p') Galois Field GF(2).

The individual vector intraframe (11 / m) elements are spatially-composed by majority scan-line memory functions of single norm #Pa1 (for 2 x 2), or a twin set ;Pax.6Paz (for 3 x 3) within the serial bi-stable transversal (;tl) memory transfer delay (see Figures 1, 2, 5 and 8).The interframe FIR transversal (#t1) with an intra-frame single interpixel (#Pa1) period embeds a spatial binary filter, with one Boolean total sum, all-or-none hierarchial (mo) vector from indistinguishable intra-frame pixel vector pairs (mi & mz) to result in the minimum single double-bound vector (mo), with a twinning definition characterisation (see Figure 2).

The outcome vector (mo) sampling probability in the divergent capturing of the 'majority' vel-correlation of the AND gate multiplier sum, with a previous-to-next recursive on the topological scan instigation cycle relationship from both the prior retrospectively preceding (m2), and the forward predictively proceeding (mi) binary vector, upon serial scan quantum ('q'-- > 'p') values within the intrafield major FIR transversal (stet) for the Galois Field GF(2).

The second spatial state-cycle race-delay again has single norm SPb1 or twin set #Pb1.#Pb2 memory delays to restore the recurrence reduced-code vector (mo) from the latency reservoir store (mo-1), in the adoptive sampling correction of the (mode) replication. The 2-pole justified percolation (mo) vector (see Figure 1) flows outside the temporal-domain (#t1) transversal processes, after the initial vector (arno) code instigation (#Pa1/2) interval within the temporal memory.

The replacement spatially bi-conjugate eigenvector (ras) of the logic window filter (LWF) is fed back into the 'doublet' transversal (sty) period. The majority intermediatary filter vector (ms) from within the Galois Field GF(2) is resultant from an applied franchise input, or scanned pixel neighbourhood, in a serial sequence with both Boolean cycle modes of a twist loop map within the temporal-domain delay.

The number of input vector ('q') elements (mt) for an indistinguishable twin or multiple code vel-correlation, with respective memory co-efficient (#Pa.) increments, for the instigation (mo) vector in the first stage, has counter balance of convergent memory co-efficient (8Pb. ) increments in the adoptive vel-selection for error correction in the second stage, to restore the uniform (ms) vector ouput quantitatively for canon class (1-2] codification in the quaternary lattice.

with reference to Figure 10, in the receiver-set the intrafield quantum 'q3' memory bi-stables (b-s) dilate the 'aut' temporal-domain (#t3) Galois pair ('q,', 'p') of the interframe delay to the recessive binary 'p3' on least frame (kn-2) for vector segments (1-4] to alternatively carry the first majority spatial (m-- > mo) filter stage within the destination intra-frame 'doublet' (see Figure 5) with appended adoptive correction, as wholly described for the vector encoder matrix. The majority spatial filter feeds both the temporal Ex-OR gate b(2/3) adaptive disjunction, and NOR gate b(4) null motion logic functions, upon the reconstruction frame (k'n).

The flag b'(n) segment monitor and the finite temporal-domain impulse response filter can similarly incorporate the majority spatial mode filter, with improved (#Pb1) adoptive reservoir correction for the vector (ms) eigenvalue functions, within the transversal of intrafield quantums ('p , The first majority (m1) vector vel-correlation circulates the intra-frame instigation, with a specific Bernoulli-Turing machine of channel equation function, and truth table relationships as conditionally follow, (mi A m2) = > mo instigation vector AND gate (#Pa1 ) m1 m2 ' & binary vectors (m1/2) O 0 0 collapsed vector 0 1 0 m2 annulled 1 0 0 m1 annulled (mo) 1 1 1 > mo sum correlation The pixel majority image from the first Boolean AND gate multiplier sum function of the logic window filter is the vel-correlation preceptor transfer (mo) vector for an inunediate synchronous output (mo), and later reservoir (mo-1) vector, with optional Median triad trees (see Figure 5 b) The second cross-correlation OR gate of serial addition results in an adoptive residual function of a stable recursive vel-selection by a specific Bernoulli-Turing machine of interchannel (mo) equation function, and truth table relationships after tree sections, as conditionally follow, (mo # mo-1) = > ms corrected vector (after triad tree OR multiplex) OR gate (#Pb1) mo mo-1 '+' binary vector (mo/o-1) 0 0 0 collapsed null vector 0 1 1 recursion add vector 1 0 1 vector code through (ins) 1 1 1 vector run code (rays) The quality code for the circuit (mi - > ms) operation of each of the instigative (#Pa1) vel-correlation AND gate function of the transfer preceptor (see Figures 1, 2, 5 & 8) is followed by the pixel memory (#Pb1) vel-selection and nested transition-adoptive OR gate transfer and characteristic of the intra-frame spatial mode majority filter (see Figures 2 and 8).The two filter-cycle program processes jointly act for the 'full-replacement' terminal (m@) eigenvector to signal the lower feed quantum ('p') of the Galois Pair ('p' R 'q') in the secondary temporal-domain (#t1). The intrinsic eigenvalues N2 = 4 are for the pixel canon classes [1-2], in successive binary intrafield (p,q) separations of a transcritical pitchfork bifurcation.

The noise and error values of a singularity noise modulation on pixel vector point are volume rejected on all capture occurrences (see Figure 7) with the limiting twinning rank adoption of a fullrcorrective (ms) elgenvector replacement.

With difficult input data the switchable option of an extended memory for the spacial multi-variable (#Pa1.#Pa2) vel-correlation for total sum vector (mo) instigations, with augmented (6Pb..Pbz) vel-selection of the double corrective adoption for the resultant (ms) vector and binary intrafield ('pz') quantum for the transition states C14] of the Galois Field GF(2) in an all-or-none response on output.

To partially overcome the permanent build-up on segment periodicity of the line-scan run-length-code due to channel error or noise with the reception of consistently degraded motion vector segmentations tl-4], the corrective (2 x 2) restoration cycle (mo - > ms) length. The cross correlation OR gate input degree (d) for vel-selection is reducible to a valency of (d-1), according to particular channel destination needs for ideal mdescriptor Cx. y,m] vector flags, when used in the fully-adapted receiver-set to reduce the excrescent lengths of segment C1-4] build-up.

The input change or motion ( 3!iq) vector (1W1 / m) gives an intrafield quantum 'qi ' binary which tracks the serial bi-stable (b-s) temporal-domain memory transfer to the intrafield binary output 'p1'. in forming the quantum partner or pair (see Figures 1, 3 and 5) in the Boolean connective dimensions of a Galois Field GF(2) mapping an (N = 22 = 4) interconnective convergence of the recursive change relationship ('p1' R 'q1') and attributes in the following neural lattice network of two co-evolving change or motion class vector C1-2] Bernoulli Turing machines Cbl-27 for conditioning m-descriptor singleton flags (see Figure 5).

Firstly for the canonical class 1 vector b(1) flags as derived from the 'fully-motive' entrant (le) image stimulus, the channel path function equation is for the quantum (+1) caricature output (m) of the quaternary lattice as follows, ('pt 'qi ') = > b(1) = > '1' = > m (green), with canon class 1 vector b(1) flags as part logic (m) for the total singleton (n,m) group, and also concurrent feed-forward of canon class 1 (m) vector b(1) flags for the temporally merger with class 6 Fallback handling for picture display.

The noise gate elimination advantage of canon class 1 vectors (see Figure 7 upper) are frame pixel controlled by the fast detection of canon class 2 vectors (see Figure 7 lower). The canon classes C1-2] have a dual correlation 2-input NOT/AND gate error noise function as follows, ((n) A b(2)) = > '1' = > ENF, or (7 Cb(1).6t23 ,\ b(2)) = > '1' = > ENF (error noise function), to simultaneously give pixel failure indication of the frame/field error noise, upon marginal irradication performance or line failure (see Figure 9) causing segment build-up.

The synchronous interframing data Cb(1).sSt23 is optionally held in further parallel (b) tertiary temporal transpose (n) delay (St2) memory (see Figure 5), for the single or multiframe period between the by-pass sampling input diversions (see Figures 1, 4, and 5), and the by-pass output re-route.

The canon class 2 motion vector b(2) flag provides a canon class 2 recurrence syncronisation lock on quin class 5 vector b(S) flags from the next matrix encoding frame (Pkn-l)/(Pkn-m) for renormalization of the class b(6) Fallback vector (green + red) to merge by motion vector two temporally-adjacent frames/fields.

to the canon class 1 vectoring (m) into the penultimate cross correlation OR gate serial sum for the class 6 vector function (see Figures 1, 3 and 5). The final hex class 6 vector resultant of the b(6) flag has vector product correspondence to associate later triad (eg Y,Cb,Cr) fetch (m = trajectories in reconstituting the class (b'n) Fallback vector of the dual canalised metric 'a fortiori' output. The hex class 6 Fallback vector as the bi-lateral b(6) flag product is a real affine collineation in an orthogonal closure identity with the input change or motion vector (M or m) on normal composite frame input, in only encoding 2-point spatial and temporal pole recurrences.

The second demolition change or motion canon class 2 vector image is the spacio-temporal resultant of circuit vectors Cb(l).Gt2 = > '1' = > n on b(2) flag synchronisation of quin class 5 vector b(5) flags (see Figures 1, 4 and 5), are from the second interframe temporal CTT3 transpose group (n,m) memory, in a parallel class 1 vector (8to) delay to compensate for initial singleton b(1) flag sampling losses omitted by the previous Boolean correlation ÇrND gate multiplier sum fuction.

The use of canon class 2 vector b(2) flags from across the FIR transversal (sty ) of the quaternary lattice provide the singleton logic control b(2) flags for integral check-code synchronisation of b(5) flags in a bi-lateral and conservative fail-safe (div n) prediction upon the vector allocation, fetching proceeding or preceding picture triad (eg Y,Cb,Cr) components for the reconstruction frame (k'n) up-conversion in the receiver-set.

kAll single pixel existential ( 3 StSq) values and odd Pixel scatter values, unique in the video framing sequence, are advantageously lost by the double sampling Boolean AND gate multiplier for canon class 1 vector b(l) flags. In both the intraframe spatial and interframe temporal regenerative filters, the duo dimensionality of image objects is fully maintained beyond only first existentials ( 3!tSq) of the 'one-bit ' pixel vector occurrence. Hence a 2-pole filter transfer 2-cycle resolution is maintained in both the duo spatial and temporal domain.

Both canon class 1 vectors and in later self-similarity canon class 2 vectors result as class 5 vectors in the b(5) flag engagement upon the interconnective synchronisation of the quantum integrated multiplier to act, under the b(2) flag controlled protocol (red) feed of the temporal transpose (n) for the total singleton sequence into the final cross correlation of the OR gate serial addition (see Figures 3 and 4). The additional double-resonance of the canon class 2 vector of the quaternary singleton b(2) flag, from the partial sum function of the dual correlation NOT/RND gate, derives from the FIR transversal.For canon class 2 vectors there is a temporal-domain anterior trajection for singleton b(2) flag synchronisation onto the supplementary canon class 2 of the temporal transpose (n) producing the class 5 vector b(5) flag (red) flow into the Boolean multiplier sum from the double correlation AND gate the quantum integrated multiplier for the 'fredkin' (see eg Zurek complexity, Entropy and the Physics of Information, Pub.Addison Wesley 1989) pixel-lock synchronisation, and a marshalling logistics as follows, [b(1).#t2 # b(2) ] = > '1' = > b(5) flag (red) so, ('p1' # 'q1').#tz # ('p2' # # 'q2') = > b(5) flag, for a simultaneous (At) total singleton (green + red) in a 'hold-back' on the coadapting network node from the temporal transpose group (n,m) between successive, or a plurality scanning of, video frames, Pk(n-l), and Pk(n-2) or Pk(n-m), or whatever, etc, on 2dimensionaltotal singleton recurrences (see Figures 1-6) for the class Fallback vector b(6) flag output resultant from the quantum b(2) integrated multiplier.

The class 6 vector Fallback b(6) flag (green + red) gives a temporal interpolative replication at destination of an images mosaic reconstruction frame (k'n) from a virtual m-descriptor value Ex,y,m] of a logic 'O', in graceful failure rebate to conservatively fetch from between the double-period, the preceding triad colour group (eg ,Cb,Cr) when an immediately proceeding frame (Pkn+m) of a logic '1' has more greatly changed data, error or noise in a way not meaningful to the content of the reconstruction frame (k'n) in a reconstitution for picture display.

The noise gate elimination advantage of canon class 1 vectors (see Figure 7 upper) are frame pixel controlled by the fast detection of canon class 2 vectors (see Figure 7 lower). 'The canon classes C1-23 have a dual correlation 2-inPut NOT/AND gate error noise function as Follows, ((n) b(2)) = > '1' = > ENF, or ( n Cb(1).1itz] A b(2)) = > '1' = > ENF (error noise function), to simultaneously give pixel failure indication of the frame error noise, upon marginal irradication performance or line failure (see Figure 9) causing segment build-up.

The synchronous interframing data Eb(1). & 2] is optionally held in further parallel (b) tertiary temporal transpose (n) delay (#t2) memory (see Figure 5), for the single or multi frame period between the by-pass sampling input diversions (see Figures 1, 4, and 5), and the by-pass output re-route.

The canon class 2 motion vector b(2) flag provides a canon class 2 recurrence syncronisation lock on quin class 5 vector b(S) flags from the next matrix encoding frame (Pkn-1)/(Pkn-m) for renormalization of the class b(6) Fallback vector (green + red).

The hex class 6 Fallback vector b(6) flag has an organizing (green + red) mapping in substitutions as follow, CCb(l) gt2 A b(2)] V b(1)] ," > b(6), Fallback vector (green + red) or as, ('p1' # 'q1').#t2 # ('p1' # # 'q1') # ('P1' # 'q1') = > b(6), class Fallback vector.

The image leading-edge of a virtual object 'covering' motion is a theoretic canon class 3 vector, the temporal dual to the anterior 'uncovering' motion of canon class 2 (see Figures 3 & 4) as realised upon the pixel termination of image object motion. The replacement quin class 5 vector for the b(S) flag having been then synchronised from both canon class 1, and simultaneously marshalled after one, or a plurality of interframe periods (Pknlm) later, under the canon class 2 image change (red) conditions (see Figures 3 & 4). The output hex class 6 vector b(6) formation (green + red = > '1') is by a universal 'vel ' double mapping and terminal convergence function (see Figures 1 and 5).

Each of the pixel vectors in a majority filter results in an associate encoder reduction of half the noise power frequency in the applicable spatial and temporal domains of annihilated single pixel value occurrences only. In simple insertion repetition within the system channel the vector encoder matrix ceases to further limit the picture image standard in maintaining the half-power spatio-temporal frequency division.

The digital signal processes (dsp) of the neural vector encoder matrix is by a serial repetition within a network distribution channel, and iteratively applied to the previously coded vector signal of the main video frame (fr/2) transmission channel without necessarily incurring any additional line quality loss, or degradation build due to pixel noise intrusion by the two canon class flb1 z reconditioning resilience of the automorphic encoded temporal (ass) and spatial (8 > a1/2) ecology definitions.

For normal video distribution, in transmission, media or recording the two encoder filters of the brpass (t T) one-bit motion vector diversion (see Figure 4) are fed alongside the binary lanes of the frame pixel (n-1)bits of chrominance (eg Cb,Cr) or luminance (Y) picture data of the signal picture (eg Y,Cb,Cr) components. The vector encoder matrix has a modular function on compression diluted bits from frame pixel-points within a distribution system.

The concurrent temporal 'cache', a serial bi-stable tT memory delay pathway for mainstream chrominance or luminance signals conveys (n-1)bits, during the by-pass interframe (d;t/2) vector encoding of a picture-frame-cycle (fr/2) duration (see Figures 2 and 4), and in a frame sequence switch to the full lane n-bit operation for full picture component conveyance when appropriate change or motion vectors are not present within the frame-lane (n-bit) signal.

With multiple-phase framing processes the vector encoder matrix extends in data storeage capacity (see Figure 1 - b) both for the interframe temporal Cb(l).sta singletons, in a frozen temporal-domain (b fits) of bi-stable holding, static and isolated from operation, on diversion from the current temporal-domain (a sot2) of dynamic bi-stable delay.

The interpolation one-bit change or motion majority vector channel gives multiple 2-dimensional processing for noise and (ElIC) intrusion volume for reliability to distribution, transmission, or recording the class b(m) vector code. The hex class 6 vector b(6) flag value has annihilated the singular vector noise and definition logic errors and failings. The particular effectiveness is upon a low event level, occasional occurrence intrusions without over-accummulating picture quality degradations.

The neural vector encoder matrix described here has image logic filter properties by the canon class C1-43 image properties in the data communications network (see Figures 1, 3 and 4) to signal code the data entropy formation. The electronic filtration device is module defined in real time and space by physical laws within canonical formulae logic for temporal and spatial pixel video formation attributes on both interframe and intra-frame scans of pixel image data. The pixel singularity rejection is on the smallest finite spacio-temporal element of a motion vector run-length-code, with only the highest pixel triad frequency considered.

For more general instrument purposes the uncompensated canon class 1 vector b(l) flag alone from the majority transversal (sot2) network may be taken as resultant matrix output, without the parallel frame sequence of the main 'cache' (A T) compensation delay.

The functional specification of the forward image change (canon class 1) and the anterior rear-edge (canon class 2) property describes for a vision or video vector a physically isomorphic and active line-scan signalling method for a hex class 6 vector b(6) flag, of an open or partitioned device embodiement concept within digital communications systems for television, video, vision, and image processing technique.

The digital vector encoder matrix (VEX) device implementation of the majority encoder logic in 1+ profile fabrication technology can use small high speed transistors and channel process differential delays of a neural homeomorphic functional method, an electronic equivalent, or similar (IGUS/CIN) computational algorithm identity.

Claims (15)

1. A method of achieving a majority vector encoder matrix using an intraframe spatial pixel recurrence filter having coded vector or flag correction, matricised within a temporal (T-D) domain (6tl) FIR transversal and multiple correlation sum function, with optional (m) secondary (6t2) feedforward in the (n.m) two-canal temporal (TT) transpose (n,m) equalisation, by canon class 2 vector (n), and flag b(2), in synchronous check-code compensation for the canon class 6 vector of the Fallback b(6) flag, to produce pixel noise volume reduction, for change or motion vector flag (b(m)) logic, for an m-descriptor, with parallel frame period ( T) delay, within television video compression systems for downconversion or up-conversion, in video distribution with frame delay compensation, for digital television, or for video, vision, or image processing from one or more Median signal sources.

2. A method as claimed in claim 1 wherein, using a two stage spatial filter with an intraframe velcorrelation capture filter is used to scan code the instigation of an adoptive vector vel-selection filter, direct and auto-correction (mO-1) output vectors (m5), for flags (p,q), or singleton flags (b(m)) are produced, giving discrete singularity noise elimination and resilience for interframe vectors within television bit-rate reduction, video or image control or measurement systems, or to remove the vector noise volume of single pixel occurrences.

3. A method as claimed in claim 1 or claim 2, wherein for an image frame signal, a reduction in the occurrence is achieved in either or both of the intraframe spatial pixel to pixel domain, and/or the interframe temporal-domain frame to frame, for quaternary class (1-4) detection, with optional secondary temporal (6t2) equalisation for double sampling compensation on pixel-point motion vectors on the change (6Pa1, etc.) duration.

4. A method of achieving a pixel majority vector encoder matrix as claimed in any of claims 1 to 3, wherein a finite (temporal-domain) impulse response or quaternary majority filter, with secondary temporal equalisation (n,m) is used in conservative graceful (dev n) compensation in frame allocation rebate by synchronising the two-canal or temporal transpose (TT) vectors (n,m) for Fallback b(6) flags of canon class 6 vectors, on video attribute change or motion, resulting in an elimination of singular vector or pixel flag occurrences in the temporal frame to frame and/or the spatial pixel to pixel vector, or in flag noise.

5. A method as claimed in any of claims 1 to 3, wherein a relative elimination of noise volume occurrences upon single pixel vectors or scan flags of an image frame sign, in both the majority spatial domain, and in the input finite temporal (T-D) domain (6t1) filter is achieved without compensation using a 'fully motive' singleton flag of Luma difference on the first matrix dual correlation sum of the canon class 1 vector as a singleton flag b(l) on the filter output for machine-vision, control and instrumentation.

6. A method as claimed in any of claims l to 5, wherein there is a singleton class b(S) vector from canon class 1 motion conditions available in multiplex diversion from a plurality of secondary temporal (TT) transpose memory (6t2) delays in by-pass synchronisation by a diversion with the main ( t T) frame (b-s) delay.

7. A method as claimed in any of claims 1 to 6, using intraframe Median hierarchial filter with multiple tree-section inputs for optional triad signal inputs.

8. A method as claimed in any of claims 1 to 7, wherein for digital image or video frame signal an mdescriptor identifier matrix is used.

9. A method as claimed in any of claims 1 to 7, wherein for a digital image or video frame signal, a decoder m-descriptor matrix is used.

10. A method as claimed in any of claims 1 to 9, wherein a motion vector (M/m) channel is used for down-conversion or up-conversion in bandwidth reduction or bit rate compression.

11. A method as claimed in any of claims 1 to 10, wherein an error noise function (ENF) indicator is used to give an indication of encoding effectiveness.

12. A majority vector encoder matrix with an intraframe spatial pixel recurrence filter having coded vector or flag correction, matricised with a temporal (T-D) domain (6tl) FIR transversal and multiple correlation sum function, with optional (m) secondary (6t2) feed-forward in the (n,m) two-canal temporal (TT) transpose (n,m) equalisation, by the canon class 2 vector (n), and flag b(2), in synchronous check-code compensation for the canon class 6 vector of the Fallback b (6) flag, producing pixel noise volume reduction, for change or motion vector flag (b(m)) logic, for an m-descriptor, with parallel frame period (AT) delay, within television video compression systems for down-conversion or upconversion, in video distribution with frame delay compensation, for digital television, or for video, vision, or image processing from one or more Median signal sources.

13. A device for implementing, within the intraframe spatial scan, an auto-corrective majority filter with an instigation (mO) logic window for full replacement (mg) vector output in the one-bit (q) temporal-domain (T-D) or discrete (M/m) transfer.

14. A device for implementing the pixel majority vector encoder matrix substantially as hereinbefore described with reference to and as illustrated in the accompanying drawings.

15. A method for implementing a pixel majority vector encoder matrix device substantially as hereinbefore described with reference to and as illustrated in the accompanying drawings.