GB2291306A

GB2291306A - Image motion flag or vector filter

Info

Publication number: GB2291306A
Application number: GB9513513A
Authority: GB
Inventors: Kenneth Stanley Jones
Original assignee: Individual
Current assignee: Individual
Priority date: 1994-07-02
Filing date: 1995-07-03
Publication date: 1996-01-17
Also published as: GB9513513D0

Abstract

Image motion flags or vectors are derived by subtracting a first image frame from the succeeding image and thresholding the absolute difference. To remove noise in the form of isolated motion flags from the resulting signal the input m is passed directly and via two pixel delays delta pa1, delta pa2 to the three inputs of an AND gate, the output of which is supplied directly and via two pixel delays delta pb1, delta pb2 to an OR gate to produce a replica of the input motion flag. If it is desired to accept image motion flags of only two pixels duration switches may set one input of the AND and OR gates to "1" and "0" respectively. Image motion may be classified into four classes - fully moving, covering, uncovering, static - by comparing the motion flags derived over two successive interframe periods; alternatively the motion class signals may provide the input to the motion flag filter. <IMAGE>

Description

Description MAJORITY VECTOR ENCODER DIRF Claims The majority vector encoder is for digital television or video systems for any purpose of signal noise volume reduction by the neural logic-filter processing of binary vector representations, as derived from pixel point interframe change values for a linear predictive motion vector upon an adjacent frame by a binary interpolation descriptor or data stream, or for one-bit per pixel-point image attribute change code detection.

The aim is to remove odd single uncoded representations of generation error, noise or interference/ENC values which need to be removed, or encoded by a digital impulse response filter (DIRF) in order to more precisely resolve, with high accuracy codifications, the maximum statistical likelihood values of pixel change or motion by a vector or flag, as data previously derived from successive kinematic video frames at the input video signal source.

The motion or change vector (m) or flag (b(m) for an image pixel is a binary quantity from a finite impulse response source to result in the single digital quantum ('qx') values from sensory vector derivation, for any quantity or quality degree of change upon the binary logic expression ('0'Z'1') from the video pixel point, as taken from the input digital television or video system signal.

There is technological adaption for use with HDTV (high definition television), MPEG (motion picture expert group) or broadcast (DVB) systems in motion vector codification at image source, and intermediary distribution positions, for the forward and rear predictive frame interpolation ( '0'), for bounded field or frame vector reconstruction (k'n) within the destination receiver set. For specific video compression systems a picture signal frame (2N T) store or bi-stable (b-s) 'cache' pixel memory is used in a frame (n-1)/n-bit lane dynamic (A T) delay for the digital picture data, from (eg Y,Cb,Cr) components, while the single binary change or motion vector is automatically diverted into the by-pass vector encoder filters.

The basis digital video methods for the majority vector encoder processes are the earlier United States Patent number 4,651,207 of Bergmann et al, and by Jones supra, Patent Applications GB 2 265 783 (A) Bandwidth reduction employing a vector channel, GB 2 265 784 (A) Video image motion classification and display, and GB 2 274 371 (A) Measurement and control using motion classification flags.

The inverse application of a pixel majority vector encoder into a digital television vector compression or bit rate reduction system results in an enhancement ability in the generation of vectors from video code (kn) reconstructions of the source majority vector encoder for channel distribution to the destination vector encoder and binary pixel interpolation of high quality picture reconstruction frames (k'n) at the destination receiver for picture screen display.

Envisaged further uses extend to encode video system measurement, control and display by motion classification flags under class canons 1-4) as for interpolation, and for pixel change detection in image processing methods.

The majority vector encoder by the specification text for device embodiement is illustrated and described by the following list of drawings.

Figure 1. is an overview diagram of a channel neural network to show the one-bit per pixel vector signal processes, in both the intra-frame spatial (6Pa/SPb) and interframe temporal transpose LTT domains (6ts or sot2), as a digital component sequence for change vectors or flags from alternate video frames in a viterbi pascal tree and mapping network sequences for vector separation fields and convolution error correction.

Figure 2. is a diagram of the intra-frame majority filter with one spatial C6Par) interpixel memory, with the optional version of a switchable extended C6Pa1.6Pa2) double memory, each within an interframe (b-s) memory delay of the temporal-domain (#t1) transpose [TT] period of the majority vector encoder (see Figures 6 b / c.d).

Figure 3. is a transition diagram for the data communications network of the image class or quaternary property model within four cascade frames [kn - kn-3] of the neural device. The class [1-4] resultant b(5) is a 2-dimensional (x, y) classification of motion engaging two canon classes Cm = 1, and m = 2 to encode the class vector b(S) output.

Figure 4. is the code sequence truth table for the pixel vector encoder from boolean gate cluster devices and neural processes, relating the image class or quaternary property model within the interframe 'cache' memory and the majority (T-D) compensation filter.

Figure 5. is a circuit diagram of neural network functions and connections of the intra-frame (#Pa1/2, #Pb1/2) spatial filter within the three frames (Pk(n), Pk(n-1), Pk(n-2)) of the temporal-domain (T-D) compensation network filter equalisation of the neural pixel vector encoder.

Figure 6 a.b / c.d. is a diagrammatic frame serial of temporal transpose TT vertical cross-sections of the change or motion vector (m) frame intrafield ('or') representations in pixel line segments Classes 1-4 taken from video frame input through the vector encoder from video frames Pkn, Pkn-1, and Pkn-2, here from source pixel data (see Figure 6 a).

Figure 7. is a comparison of the noise volume performance illustrated between the correlation b(1) vector sum for the class 1 flags of Luma difference occurrences in reference against the differential correlation b(2) dual vector for the class 2 flags of Meana difference occurrences, each relating to the frame cross-sections of Figure 6 a.b / c.d, shown here without the pixel majority spatial filter.

The majority vector encoder (see Figures 1 and 5) scheme is the synchronous neural combination for each of the temporal ( & z) and the spatial (SPai) operations of principally two separate image filter transfer functions. Each vector filter section operates upon a common interframe change or motion vector from input detection !6q = > m, for 'qi']. The primary intra-frame representatives from the previous (^ P) FIR filter vectors (m) result in the neural 'qi binaries being recoded by a convolutional (q,p) viterbi tree circuit (see Figures 1, 2 and 5), to form the Galois Field GF(2) (with n2 = 4 pixel canon classes) in successive (p,q) frame separation.The Galois Field GF(2) occurrence is during the major interframe of the temporal-transpose ETT] domain interval of either Hamilton's quaternary (#t1), or secondary canon class 1 singleton in recession to the quantum delay vector (b(1).8t2). The intrafield vector input 'qi' binary to the bi-stable memory of the first quaternary delay dilates later into the forward feed as the interleave binary output 'p1' quantum (see Figures 1, 5 and 6 b) for the neural lattice class mapping.

To facilitate the secondary image dilation property with the latertemporal (6t2) development in the singleton b(1) multiplier correlation of the AND gate sum function (see Figures 1 to 5) supplements (gut) transpose canon class 1 vectors b(1) to the dilation lattice differential correlation of the NOT/AND gate dual function. The pixel synchronous neural engagement of the correction onset [b(1).#te # b(2)] mapping upon canon class 2 vectors equalises the needed compensation from the initial sample sum losses attributable to the canon class 1 singleton vector alone (see Figures 1, 5, and 6 a/b). The first convolution (m) correction mapping is for a change or motion class 1 conditions, by a boolean multiplier from an SND gate sum correlation.The double finite correlation sum is sampled twice across the entire interframe memory delay (#t1) period. The individual interframe (#Pa1, #Pa2) bi-stable (#t1) delays overall has multiple interpixel memory elements (8Pal, sPa2,..., Span, SPam., see Figures 1, 2 and 5). The interframe intra-frame of interpixel (6Pai, #Pa2) delay elements as the majority (#Pa1) single or (6Par.6Pa2) double period spatial filter, with the statistical mean probability of the next data sampling having the AND gate auto-correlation relationship from both the retrospectively preceding, and the predictively proceeding binary value.The later memory delay (Pb1 or 6Pb.6Pb2) elements of the auto-selection adoptive correction follow after the vector code instigation, for the first interpixel (6Pai or sPaz.6Pa2) memory auto-correlation cycle, and for the substitute output vector (o) of the logic window filter. The interpixel memory (#Pa1, #Pa2) is additional to the interframe (#t1) time period elements (#Pan, #Pam, etc.) without increasing the overall (#t1) delay interval of the Hamilton bi-stable transpose.

The receiver set interframe quantum 'q3' memory (b-s) bi-stables in the temporal (#t3) transpose delay to binary 'p3' can optionally carry the intra-frame majority spatial filter circuits in the destination system (see Figure 5).

For the flag b'(n) segment monitor the finite (temporal-domain) impulse response filter can optionally incorprate the pixel majority spatial filter with or without interframe (T-D) compensation delay filtering.

The first majority vector auto-correlation circulates the intra-frame pixel check code for intra-frame instigation with a specific machine truth table as follows,

(#Pa1) q1 q2 ' & o 0 0 o 1 0 1 0 0 (mi) 1 1 1 AND gate.

binary vector (6) collapsed annulled annulled sum-correlation The pixel majority image from the first AND gate function of the logic window filter is the auto-correlation preceptor transfer for the second OR gate corrective auto-selection (o) function and an output section vector resutant. The second OR gate vector adoptive function of auto-selection has a specific machine truth table as follows,

(#Pb1) m1 m2 '+' o 0 0 o 1 1 1 0 1 (n) 1 1 1 OR gate binary vector collapsed vector recursion vector vector code through vector run code (o) The quality check code for circuit (6 - > o) operation of each of the auto-correlation (#Pa1) and instigation AND gate function of the transfer preceptor (see Figures 1, 2 & 5) is followed by the memory (SPbi) auto-selection and transition adoptive OR gate transfer characteristic of the intra-frame spatial mode majority filter (see Figure 2). The two majority filter cycle processes jointly act to signal (o) the lower forward feed data (n) of the temporal transpose group (n, m) to deplete the noise and error values by the volume rejection ratio of the canon class 1 vector capture occurrences (see Figure 7).

With difficult input data the receiver set has the option of an extra spatial interpixel (sPa9) auto-correlation for vector (mi) instigation, with subsequently a broader auto-selection (o) corrective adoption (n) for the class b(S) vector output.

n second change or motion class 2 vector channel is resultant of circuit vectors b(2) (see Figures 1 and 5) from the interframe differential dual correlation of the boolean 2-input AND/NOT gate cluster function. The rear or anterior image uncovering of canon class 2 motion results in the protocol of each vector with the meana signal, error and noise occurence characteristic. The heuristic two vector protocol Cb(1) c/w b(2)] is a temporal 2-frame parallel of the double finite canon class 1 vectors, from the first interframe (sty) period with those of the second interframe ( & 2) period.

The quantum integrated multiplier coupling with canon class 2 vectors is only with the first interframe (tor) period 'qi' binaries, both the image vector singletons are in pixel-point frame synchronisation.

The motion class 2 vector alone results in singleton vectors indicative of scan line pixel sampling failure to produce singletons across the video frame or field (see Figures 5 and 7).

The probable occurrence of odd noise speckle volume or quantisation error in singularity within the differential dual correlation vector signals of canon class 2 is overcome from the multiple correlation SND gate sum function input to the third 2-input correlation Cb(l).t2 fas b(2)] multiplier sum of the NND gate function, with pixel vector synchronisation when and only when, (b(1).t2 A b(2)] = > '1', is in simultaneous concurrence for a network nodal analysis between successive frames Pk(n-1) and Pk(n-2).

The interframe one-bit binary data is held in the secondary interframe (b-s) memory (8t2) delay for the frame period between the by-pass sampling input diversions (see Figures 1, 4 and 5) and the by-pass output diversion.

The third space and time vector samples are the first pixel [b(1) from ( & i)] resultant signal feeds into the final cross-correlation OR gate multiplier input. The precise pixel point registration of vector concurrence with the subsequent pixel b(2) vector, from the immediately successive (sty) interframe, enables from the second temporal transpose ETT] an engagement group rb(1).St A b(2)] supplementary vector by the 2-input OR gate function of the quantum integrated multiplier. The delay ( & 2) quantum integrated vector (see figures 5 and 6 b/c) compensates for the initial double finite sampling loss incurred of the class 1 pixel conditions.

The supplementary vector equalises the compensation Eb(1). & 2 A b(2)] encoding with delayed convolution sampling period correction. The multiple correlation b(1) sum function upon spatial majority filter samples (n) forming the first temporal transpose group (n, m) results in a high degree of singular noise volume rejection. The correlation sum vector from the interframe memory delay (b(1).6tz) results in the additional control of the quantum integrated multiplier by the subsequent b(2) vector to equalise the b(1) sampling loss compensation for a substitution by the class b(5) replacement vector for b'(n) as follows, C[b(1).t2 Ab(2)] V b(1)] = > b(5), at frame Pkn-2.

Each of the pixel vector filter results in an associate encoder reduction of half the noise power frequency in the applicable domain for single pixel value occurrences only. In repetition the majority vector encoder ceases to further degrade the picture image standard beyond the half power frequency division.

In normal combination the majority vector encoder exhibits a standard truth table sequence (see Figure 4) with a binary boolean characteristic (see Figure 3).

The digital signal processes (dsp) of the vector encoder in repetition can be iteratively aPPlied to the previously coded video signal of the main frame (fr/2) transmission hannel without necessarily incurring any further degradation loss encoded temporal (sty/2) or spatial (sPai) definition.

For normal video distribution, in transmission, media or recording the two encoder filters of the by-pass (iST) 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) elements. The encoder may use diluted frame pixels except at source.

The concurrent temporal 'cache' bi-stable 4 T memory delay pathway for mainstream chrominance or luminance signals conveys (n-1)bits, during the by-pass interframe (sty) vector encoding of a picture-frame-cycle (fr/2) duration (see Figures 4 and 5), and in 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 signal.

The interpolation one-bit change or motion majority vector encoder channel gives multiple two dimensional processing for noise and (EMC) intrusion volume resilience to distribution, transmission, or recording the class (m) vector code. The class vector b(5) value replaces the singular noise and definition logic error failings.

The theoretic leading image impulse response b(3) vector for canon 3 class change is a virtual vector dual to the uncovering canon class 2 (see Figures 3 & 4) resultant in the substitution class vector by the second temporal transpose (6t2) group and the supplementary engagement vector Cb(1). & 2 A b(2)] for the canon class b(5) vector produced in realisation upon the pixels termination of image motion. The substitute replacement having been derived from both canon class 1 and one interframe period later the canon class 2 image change conditions (see Figures 3 & 4).

The neural majority vector encoder described here has image logic-filter properties by the boolean image property in the communications network to signal code the data entropy formation. The time and space phenomenon device is held in physical law by temporal and spatial pixel point video attributes of interframe and intra-frame image data. The description specifies any physically isomorphic and active scan line signalling method for open embodiement within digital communication systems concepts for television, video, image processing and machine-vision technique.

The digital device implementation of the pixel majority vector encoder logic is in 1+ improved integrated technology or any neural homeomorphic functional identity.

Claims

Clai

1. The device of a pixel majority vector encoder in an intraframe spatial concurrence filter transfer with correction which may be within a temporal-domain correlation sum function with compensation transfer, with one or either resultant in noise volume reduction, for binary vector logic within television video compression systems for down-conversion or up-conversion, in signal distribution, or for digital image processing.

2. n method for a majority vector encoder as claimed in claim 1, wherein a two stage intraframe auto-correlation attribute filter code instigates an auto-selection adoptive correction output vector or flag, for resilience within television bit rate reduction, video or image control or measurement systems to remove single pixel noise volume occurrences.

3. n method for a pixel majority vector encoder as claimed in claims 1 and two, wherein there is an interframe finite temporal- domain quaternary filter with compensation for attribute change or motion vectors, being resultant in a lowering of single pixel occurrences in the noise volumes.

4. n majority vector encoder as claimed in claims 1 to 3, wherein for a video or image data frame signal there is a relative elimination of single incidence noise volumes in either or both of the intraframe spatial domain, and/or interframe temporal domain with quaternary detection equalisation for the change or motion duration.

5. R pixel majority vector encoder device substantially as hereinbefore described with reference to and as illustrated in the accompanying drawings.

6. n method for a pixel majority vector encoder substantially as hereinbefore described.