JP.2005-253015,A
`
`Bibliography
`(19) [Publication country/region! JP
`(12) [Kindofoffirial gazette] A
`(LD [Publication number! 2005253015
`
`{43} [Dateof publication of application] 20050015
`(54) [Title of the invention] APPARATUS AND METHOD FOR DETECTING MOTION VECTOR,
`
`AND PROGRAM
`
`(50) Unternationsl Patent Classification Teh Edition!
`
`ROaN 7/82
`
`[FT
`
`HOGN Fee
`
`za
`
`Request for examination] Unrequested
`Numberof claims! 15
`
`[ [ [
`
`Format of applieation} OL
`me
`[fotal munberof pages! 49
`(21) [Application nuanber} 2004064717
`(22) [Filing date] 26040308
`(71) [Applicant]
`[Name] MATSUSHITA ELECTRIC IND CoOLTD
`
` F-term (veference)|
`
`(74) (Representative!
`Identification reunber!] 16G090446
`
`Patent attorney!
`INamel NAKAdIMA, Shiro
`
`(72) Unventor!
`
`Fall name} NISHIOKA SHINICHIRG
`
`Theme code (reference)]
`
`SCGS59
`
`SCOSGLBIS
`
`BCOSOMAG
`
`BCDS OMAGS
`
`SCOSAMA TA
`
`SCOSOMALS
`
`SCOsEMCTL
`
`SCOSONNOS
`
`BCOSON NOD
`
`

`

`SCOSONNIS
`
`SCOSPPOS
`
`SCOSSPPOS
`
`ACOSSPPOT
`
`ACOSOTAGZ
`
`aC9TROS
`
`SOGSATCNZ
`
`SCOSUTOUR
`
`SCOSSTCOS
`
`SCHSOTCAS
`
`ACOSATDOS
`
`aCoseTDo?y
`
`SCOOASTD2
`
`SCOTTAGS
`
`
`
`(S7) [Ovexsvsiew!)
`
`
`PROBLEM TO BE SOLVED
`
`
`To reduce the competation
`
`8
`
`amore required in the motion vector detection with
`
`
`
`oreo
`
`pinels.
`
`SOLOTION: There are provided a difference evahition
`
`value computing unit for outputting a difference feature
`
`amount containing positive and negative signa of the
`
`difference,
`
`together with a difference evaluation value
`
`SAD: a difference feature ammuit comparing unit for
`
`campering the difference
`
`feature amounts bebween
`
`adjacent reference blocks: and a completiondecide nit
`
`precision which is lower than the integral number af
`
`
`
`for deciding the end of the motion vector detection,FFgehyieeeof
`
`depending on the difference evaluation value SAD ard the
`
`result of the difference feature amount comparison, arc
`
`high precision motion vector detection is carmed out be
`using the difference feature value generated, when the
`
`motion veetor with the precision of an integral number of pixels ig detected.
`
`
`
`
`

`

`* NOTICE *
`
`JPO and INPIT are not responsible for any damages canaed by the use of this translation.
`
`1. Phis document has been transiated by computer. Se the translation may not reflect the original
`
`precisely.
`
`3, **** shows a word which cannot be translated,
`
`3. In the drawings, any words are not translated.
`
`Seope of Claims
`[Patent Claims!
`
`[Cham #1
`
`After a plurality of reference tlorks are set stepwise in the veference image, a | optimum block
`
`having the highest similarity with the current bleck in the encoding target image is searched from
`
`the seb reference block.
`
`in the motion veetar detecting device, one oy more interpolation blacks obtained by imterpolating
`
`pixels bebween reference blocks are addedto a comparison object to obtain a 2 optimum block having
`
`the highest similarity, and a motion vector is deteeted with an accuracy of an integer pixel oy less
`
`based on the 2 optimum block,
`
`Far each of the set reference blocks, a caleulation means for calenlating a difference characteriatic
`
`anionit Indicnting a sion ofa difference betoreen the carvespornding pisel of the eurrent block andthe
`
`corresponding pixel of the reference block for each pixel in the block ;
`
`Comparing means for outputting a comparizon result showing a possibility that @ similartty of an
`
`interpolation block obtained by interpolating pixels between the adjacent reference blocks is higher
`
`than a similarity of any of the adjacent reference blocks by comparing the difference feature amounts
`
`caleniated far the adjacent reference blocks }
`
`Adetermination means for determining that the search for the | optinrum block is terminated when
`
`the comparison result indicates a predetermined possibilty :
`
`Amotion vecter detection aqmearatus comprisiag °.
`
`[Nain g
`
`The calculation means calculates a difference feature amount indicating only the positive and
`
`negative signs of the difference of the corresponding pixel fox each pixed in the block.
`
`The comparison means vutputs the number of pixels whose positive and negative signs indicated by
`
`the diference feature amount do not match as the comparison result.
`
`When the conrparison resuli is equal
`
`to or larger than a predetermined threshokl value, the
`
`determination means determines that the search for the 1 optiroal block is terminated,
`
`The motion veeter detection device according to claim 1 chereeterizad by things.
`
`[Claim 3]
`
`In addition to the positive and negative signs of the difference of the corresponding pixels, the
`
`calculation means calculates a difference feature amount indicating a quantization value of an
`
`

`

`absolute amount of the difference for each pixel in the block.
`
`The comparison means outputs, as the result of the comparison, the numberof pixels whose positive
`
`and negative signa indicated by the difference feature amount do not match and whose quantization
`
`valgea match.
`
`When the comparison result is equal
`
`to or larger than a predetermined thrashold walue, the
`
`determination means determines that the search for the | optimal block is terminated.
`
`The motion vector detection device according to claim1 characterized by things.
`[Claim 4]
`After a plarality of veference blocks are se stepwise in the reference image, a i optimum bleck
`hueving the highest sirniarity with the current bleck in the encoding target umage is searched from
`
`the set reference block.
`
`Tn the motion veetor detection method, one or more interpolation blocks obtained by pixel
`
`interpolation betwenn reference blacks are addedto a roraparison object to abiam a 2 optinnon block
`
`having the highest similarity, and a motion vector is detected with an acsuracy of an uitegerx pixel or
`
`less based on the 2 optimumblonk,
`
`Fer each of the set reference blacks, an arithmetic step for calculating a differential characteristic
`
`quantity indiesting a sien of a difference betweena corresponding pbxel of the current block anda
`
`corresponding pixel of the reference block for each pixel in the block :
`
`An estimation step of estimating a degree of similarity of an ieterpolation blork obtained by
`
`interpolating pixel between the adjacent reference blocks rather than the similarity of the adjacent
`
`reference blocka by comparing the calculated difference features with each other.
`
`Determining that
`
`the search of
`
`the 1 optinvurm bleck is terminated if the determination is
`
`affirmative ;
`
`A mation veetar detection method eamprising °.
`
`[Claim 5]
`
`To obtain the 2 optimum block by adding only an interpolation block posttively estimated m the
`
`estimation step tio the comparison obhect.
`
`A motion veetor detection method according ta claim 4, wherein said motion vector is a motion
`verfon
`
`[Claim6]
`
`Af ex a Lo aptiouan bleck having the Inghest sindlaxity with @ current block ro an encoding target
`
`poage is searched ont of a phurality of reference blanks set tp a reference image, a first optimal block
`
`is searched.
`
`In the mation veetar detection method, one ar mere interpolation bincka obtained by pixel
`
`inferpalation between reference blacks swe added to a comparisonobject to obtain a 2 optinvam block
`
`having the highest similarity, and a motion vector is detected with an accuracyof ar integer pixel or
`
`less based on the 2 optimumblock.
`
`

`

`Fer each of the reference blocks, an artthmetic step for caleulating a differential characteristic
`
`quantity showing a positive / negative sign of a differance between the corresponding pixel of the
`
`current block and the reference block for each pixel in the block :
`
`An astimation atep of estimating « degree of similanty of an interpolation block obtained by
`
`iiterpolading pisel between the ackacent reference blocks rather than. the similarity af the adineent
`
`reference blocks by comparing the calculated difference features with each other
`
`Implication,
`
`Yo obtain the 2 aptienan black by adding only the interpolation block positively astimated in the
`
`estimation step to the comparison object.
`
`Axsetion vector detecting method characterized by things.
`
`[aim 7]
`
`Tn the caleulation step, a difference feature amount indicating only the positive and negative signsof
`
`the difference of the corresponding pixel ia caleuated for each picel in the bleck.
`
`When the number of pixals in ewhich the sign of the sign indicated by the difference feature valueis
`
`not coincident is equal to oy larger than a predeternrined thresheld value, the estimation step is
`
`positively eatimatert.
`
`Amotion veetor detection method according to any one of cladma 4 to 6.
`[Claim 8]
`
`In addition to the positive / negative sien of the difference of the rorvespardingpisel, the calculation
`
`step calculates a difference feature amount indicating a quantization value of an absolute amount of
`
`the diference for each pixel in the block.
`
`The estimation step is positively estimated when the number of pixels whose positive and negative
`
`signs indicated by the differential feature amount do net match and whose quantized values coincide
`
`with each ather is equal to or larger then a predetermimed threshold value.
`
`Amotion vector detection method according to any one of claims 4 to 8,
`
`[Claim ol
`
`Al reference bleck and a 3 reference block which are set in a phirality of times im the 1 reference
`
`unage and the 2 reference image which are displayed beftye and after the time af the image to be
`
`encoded, and one or more time Interpolation blocks which are obtained by pixel interpolation
`betweenthe | reference block andthe 2 reference block ;
`
`In the motion vectur detecting method, an optinrum block having the highest sinilarity with a
`
`carrent block bathe image te he eneoded is cbtained and a motion vector ts detected birectionally
`
`baserl on the optimum block.
`
`A tL optimumblock baving the highest similarity fromthe | reference block, and a l calaidstion step
`
`for caleulating a i differantial characteriatic quantity for indicating a positive f negative sign af ¢
`
`difference between the curvent block and the corresponding pixel of the | aytimum blockfor all of the
`
`pixels of the block ;
`
`Aasearch step for searching a 2 optumam bleck having the highest similarity fram the set 2 reference
`
`

`

`bloek while setting the 2 reference block in a stepwise manner}
`
`For seach of the set of 2 reference blocks, a 2 calculation step for calculating a 2 differential
`
`characteristic qusortity for indicating the positive and negative signs of the difference between the
`
`correaponding pixels of the current block and the 2 reference block for all of the pixels of the block :
`
`An estimation step of estimating, by comparing the 1 difference feature quantity and the 2 difference
`
`feature quantiby, whether or nol a similarity of a time interpolation block obtained by interpolating
`
`pixel between the 1 optiraam block and the 2 reference bloek je Iigher than that of the first and
`
`second reference blocks :
`
`Determining that
`affirmative 5
`
`the search ef the 2 eptimum block is terminated if the determination is
`
`A motion vector detection method comprising<.
`
`[Claim id
`
`in the 1 eaiculation step, a 1 differance feature amount indicating only the positive and negative
`
`signa of the corresponding pixel for eachpixel in the blockis calculated.
`
`Tn the 2 caleulation step, a 2 difference featars amount indicatimg only the positive and negative
`
`sizns of the corresponding pixel for each pixel in the block is calculated.
`
`Tn the ease where the number of the positive { negative signa indicated hy the 1 differential feature
`
`amorott and the 2 differesitial feature amcrort is equal to or larger than a predetermined threshold
`
`value, the estimation step is poaihively estimated.
`
`A motion vector detection method according to claum 9, wherein said motion vector is a motion
`yerkor.
`
`[Qlaim1]
`
`In the ft ealendation step,
`
`in addition to the positive { negative sign of the difference of the
`
`oorrespording pixel, a 1 difference characteristic vahie is caleulated for each pixel in the block.
`
`In the 2 calculation step,
`
`in addition to the positive / negative sign of the difference of the
`
`corresponding pixel, a d difference characteristic vahes is calculated for each pixel in the block.
`
`in the case where the sign of the sign indicated by the 1 difference feature value and the sign of the 2
`
`difference feature value are not coincident and the wamber of the pixels whose quantization values
`
`cohncide with each other js epaal ta or larger than a predeterminer threshold valu, an affirmative
`
`estinuahon is made.
`
`A rotten vector detection method aceording to claue 9, wherein said raction vector is a aiotion
`Vechor,
`
`[Claim 1]
`
`After a l optumum biock having the highest similarity with a current block in an encoding target
`
`image is searcher out of a plorality of reference blocks ect in a referexce image, a first optimal block
`
`33 searched.
`
`The computer executable program is a computer executable program for obtaining a 2 optimumblock
`
`

`

`having the highest similarity by adding one er more prerpolation blocks obtained by pixel
`
`interpolation batwean reference blocks to a comparison object, and detecting a aintien vector with an.
`
`aceuracy of an integer pixel or less based on the 2 aptivnumblock.
`
`Acomputer causes & computer to execute the steps af at least one of chums 4 to 8.
`
`A program cbararterized by things.
`[Naim 13!
`
`Al reference block and a 2 reference block which are sat in a plurality of times in the 1 reference
`
`image axthe 2 reference immuwhich are displayed before and aller the me of the image to be
`
`eneoded, and one or more time interpolation blocks wiueh are obtained by pixel interpolation
`
`between the | reference blockand the 2 reference block;
`
`This computer program is a computer executable program for finding an aptinvem block baving the
`
`highest similarity with a current block in an image to be encoded and detecting a motion veetor
`
`bidirectionally based on the optizaam black.
`
`A computer canieas 4 computer fo execute the steps of at least one of claima 9 ta U1.
`
`A program characterized by things.
`
`Detailed Description
`(Detailed description of the Invention!
`‘Technical field!
`fooad]
`The present inventionrelates to an apparatus, a method and a programfor performing
`motion vector detection, which is used for video compression coding or the like, with an
`aceuraey of an integer pixel or less, and more particularly, to a technique for reducing a
`calculation amountrequired for motion vector detection.
`{Background of the Invention]
`looo2]
`Moving picture data generally has a property that an image of a frame and an image ofa
`next frame are similar,Although this is referred to as redundancyin the time direction,
`it is possible to greatly reduce the overall amount of data as compared with a system in
`which pixel data of each frame is directly stored or transmitted by taking a difference
`between frames.
`- 2, and - 4 which are the international
`MPEG (Moving Picture Experts Group) - 1,
`standard of the eading method of moving picture eloment data;
`Image compression technique which combines DCT (Discrete Cosine Transformation }
`conversion and motion compensation (Motion Compensation), and to reduce the
`tenrporal redundancy of the motion picture, to reduce inter-frame prediction and motion
`compensation (‘Motion Compensatio)
`
`

`

`tn addition, a large amountof moving picture clement data is efficiently compressed.
`{a008]
`
`Motion compensation is also referred to as motion vector correction, and a motion vector
`(Motion Vector) is detected as a vector quantity indicating the magnitude and direction
`of motion of an object moving on a screen by comparing pixel data of 2 consecutive
`frames in the time axis direction, and this motion vector is encoded as a difference
`
`between frames. In the detection of the mation vector MV. first, a mohion vector MV with
`aninteger pixel precision is searched, and then a motion vectorMV 2 with a half pixel
`precisionis searched,
`flood]
`(integer Pixel Precision Motion Veetor Detection)
`FIG. 10 is a diagram showing an overview of motion vector detection. As shown in FIG. 4,
`in detection of a motion vector, first, a block (referrad to as a current block) is selected
`from among the encoded image [<SUBT</SUB> (currentframe) at time T divided into
`blocks of 16 X 16 pixels, and the coordinates of the upper left end of the block are set as
`(x, y).
`[0008]
`Next, a predetermined search range of size SRxxSRy (block unit) set around the
`coordinates (x, v) of the current block is set every time one motion amount(dy, dy) of 1
`pixel units is givenboth im the X direction and in the Y direction in the reference image I
`L, The coordinate of the upper left end is (x dx, y dy), and one region of the same size as
`the current block is determined, whichis referred ta as a reference black.
`[a008]
`While gradually changing the amount of motion (dix, dy), the image pattern of the
`yeference block is compared with the image pattern of the current block, and the degree
`of similarity is determined.
`The absolute difference sum SAD (Samof Absolute Difference), the difference square
`
`sum SSD (Sum of Square Dif srence) shown im expression (8), ete. which say the
`quantity for measuring the similarity of two image patterns as a difference evaluation
`value, and are shown inthe helow-mentioned expression(7) are used.
`foog7|
`A result of comparing the similarity of an image pattern of each of a plurality of
`reference blocks with an image pattern of a current block is obtained. A reference block
`faptimum. reference block MBK) having the highest similarity is selected, and the
`motion amount (dx, dy) of the seleeted optimum reference block MBK is set as the
`motion vector MV with an integer pixel precisionfor the current block.
`After detecting the motion vector MV = (dx, dy) with an integer pixel precision for all
`
`

`

`current bloeks in the current frame, a motion vector search with a halfpisel precision is
`perfermed in which accuracy is improved te half pixel in order to further improve
`
`accuracy.
`locos!
`(half-pixel precision motion vector search)
`Ina halfpixel precision motion vectar search, a “half-pixel precision reference block" is
`ereated by interpolation processing of pixel data. This is achieved by performing
`interpolation Ginear interpolation) of pixel data on 2 adjacent reference blocks around
`the reference block BRK Raving high image similarity corresponding to the detected.
`
`motion vector MV. By generating a half-pixel precision reference block HBK whichis
`located between the half-pixel precision and the half-pixel precision reference binck
`THBK and comparing the image patterns with the current block, a reference block having
`
`a higher image similarity than a reference block with an imteger-pixel precision is
`searched for, and a motion vector MY 2 with a halfpixel precision is detected.
`foous]
`FIG. 27 is a diagram showing an outline of interpolation processing of pixel data in the
`detaction of a motion vector MV 2 having a halfpixel precision. In the vieinity of the
`optimeam reference block MBK corresponding to the mimimum value of the difference
`avaluation value SAD obtained by detecting the motion vector MY with an integer pixel!
`precision, the pixel data is interpolated te the midpoint of 2 adjacent pixels as shown in
`the drawing, thereby generating a reference image with a balfpixel precision.
`10610]
`FIG. 28 is a diagram showing a search position in a halfpixel precision motion vector
`detection. The ***
`
`As described above, firat, (dx 2, dy 2) = {¢ 1 / 2,1 / 2)) Gx 0. dy 2) = G+ 1 / 2/ 2h,
`respectively,
`is obtained from the pixel P 1 at the upper left end of the optimum
`reference blockMBBK. By setting (6,1/2), G / 2,1 / 2), & 1/2,0), (12,0), & 1/2, - 1/2), (0, -
`V2), G/2,
`- 2h, 8 half-pixel precision reference blocks HBK 1 to HBK §, which are
`shifted by one half pixel each, and whose 8 points P 1 te P 8 are the woperleft edge, are
`generated.
`foo11]
`FIG, 29 is a diagram showing the positional relationship and the overlap state of the 8
`halfpixel precision reference blocks HBK 1 ta HBK8 generated ag described above. As
`shown in FIG. 3, most of the half pixel blocks HBEK 1 te HBR 8 overlap with the other
`half pixel blocks.
`In order to generate a reference block with a half-pixel precision, pixel data at a position
`of a half-pixel
`is interpolated from each pixel data of a reference block with an
`
`

`

`integer-pixel precision. For example, when pixel data of 5 points,, f g, h, and I at
`positions of half pixel in an area surrounded by these 4 pomts is generated from pixel
`data of 4 pomts, A, B, C, and D shown in FIG. 27, pixel datais linearly interpolated ag in
`the following equations (1) to (5). In the following equation, / / represents an averageof
`rounding,
`0012]
`
`(Mathematical formula U
`
`ee{AGB) £2
`
`[0013]
`
`iMathematical formula 2!
`
`=(A+C) //2
`
`fooid]
`
`iMathematical formula3]
`
`g=(A+B4+C+D) 774
`
`fois]
`
`iMathematical formula 4]
`
`h=(B+D)} //2
`
`0016!
`
`iMathematical formula 5]
`
`is(C+D) //2
`
`

`

`As described above, by fivst interpolating oixel data of integer pixels and generating
`respective pixel data at a half pixel position, the referenee block NBK 1 to HBK 8of half
`pixel precision, which ia obtaimed by shifting the optimum referenee block MBK byhalf
`pixel, is generated. Next, as with the 8 halfpixel precision reference blocks HBK 1 to
`HBK 8, the difference evaluation values SAD 1 te SAD 8 are ragpectively caleulated as
`in the case of an integer pixel precision, and compared with the minimum value of the
`SADcorresponding to the optimum reference block MBE.
`{00171
`According to the above comyparison, among the 8 halfpixel precision reference blocks
`HBE 1 to HBE&, the difference evaluation value 5S is smaller than the minimum value
`of the SAD corresponding to the optimumreference block MBK.
`
`If there is a reference binck which gives AD, the reference block is a block which gives a
`minimum valee of SAD, and a motion vector (ds, dy) (dx 2. dy 2) = (dx 3, dy 3)
`corresponding ta the reference block is set as a motion vecter MV 2 of half pixel
`precision,
`As described above, in order to detect the motion vector MV 2 with halfpixel accuracy
`the reference blocks HBEK 1 to HBK 8 with halfpixel accuracy are generated by linear
`interpolation, and further, the SAD calculation of these is required 8 times total. On the
`other hand, considering that the minimamnumber of times of SAD caleulation in the
`motion search with an integer pixel precision using a screen shown in FIG. 11 bis a total
`of 5 times in PO to P 4, the detection of the motion vectors MV 2 with half-pixel accuracy
`requires an arithmetic amount that cannot be ignorad. *S****5-8%4-5_
`looigl
`
`However, even if the motion vector search with the half-pixel precision described above
`is performed, the motion vector MV 2 with a halfpixel accuracy is not necessarily
`obtained at the half-pixel position. For the & half+pixel precision reference blocks HBK ]
`to HBR & which are generated by interpolating pixel data, the difference evaluation
`values SAD Ito SAD 8 are calculated and the optimumreference block MBK is selected.
`As a result of comparison with the minimum value of SAD with an integer pixel
`precision, S halfpixel precision reference blocks HBK 1 to HEE S are obtained. When
`there is no reference block which gives the difference evaluation value SAD which is
`smatier than the minimum value of the SAD withthe integer pixel precision, the motion
`veetor MY= (és, dy) with the original integer pixel accuracy becomes the motion vector
`MV2 with the half pixel accuracy as it is. In this case, calculation processing spent in
`searching for halftpixel precision is wasteful.
`loo19l
`
`

`

`te estimate whether a smaller difference
`is suffielent
`it
`fo prevent such waste,
`avaluation value SAD can be obtamed at a halfpixel position before performing mation
`detection of halfpixel precision. For this purpose, using the results of the motion vector
`detection with an integer pixel precision, the difference absolute sur aum SAD 0 of the
`position of P 6 shown in FIG. 11 a and the average of the sum of absclute differences
`SAD 1-SAD 8 of the positions of the peripheral integer pixels P 1 - P 8 of the periphery,
`As @ SATH) / 8 are taken, and the absolute value of A = G) SAD) / 0 is obtained.
`ACBADOILA technique is disclosed in the following patent document in which a amaller
`difference evaluation value SADis obtained at a half pixel position, and ¢ motion veetor
`detection is performed with a halfpixel accuracy, only when a predetermined threshold
`value is larger than &.
`[Patent document ilJapanese Patent Laid-Open No. 278696 2000
`{Disclosure of invention!
`[Problem to be solved by the invention!
`foo20]
`However, according to the above-mentioned prior art, although the possibility of
`obtaining a smaller difference evaluation value SAD at a halfpixel pasition arcunda
`center position is collectively estimated, it is net possible to accurately knowindividual
`possibilities at individual halfpixel positions.
`Therefore, it is difficult to reduce the amount of caleulation by searching for a motion
`vector only at some halfpixel positions which are more lkelyto be possible.
`loo2
`
`Further, in the prior art, after obtaining a motion vector with aninteger pixel precision,
`a half pixel position indicating a smaller difference evaluation value is ohbtammerd around
`the motion vector, In particular, when the motion vector of the integer pixel precision is
`obtaimed by the local search, evenif there is a half pixel position indicating a preferable
`difference evaluation value (e.g., a minimumvalue) in the middle of the search path, it
`ig ignored.
`Therefore, not only the preferred motion vector is missed but alse a local search cannot
`he terminated at an early stage, resulting in an increase in calculation amount.
`foog2.
`
`to salve the
`invention will be made in order
`the present
`A possibility that
`abover- mentioned problem, and a amaller difference evaluation value will be obtainedin
`each half a pivel position -- a half a pixel position-- presuming individually and with
`high precision, It is an object of the present invention to provide a motion vector
`detecting apparatus, a motion vector detecting methad, and a program for detecting a
`motion vector having an accuracy equal to or smaller than an integer pixel with a
`smaller caleulation amount than a conventional one by determiming a newconditionfor
`
`

`

`terminating a focal search earher than a conventional one.
`
`iMeans for solving the problem]
`foozal
`In order te achieve the aboveobject, a motion vector detection apparatus of the present
`invention searches for a Ll eptimal block having the highest similarity with a current
`block in an eroding target image from among a set reference block, while setting a
`plurality of reference blocks in a reference image. ln the motion vector detecting device,
`one or more interpolation blocks obtained by interpolating pixels between reference
`blocks are added to a comparison object to obtain a 2 eptimum block having the highest.
`similarity, and a motion vector is detected with an accuracy of an integer pixel or less
`based on the 2 optimumblock. For each of the set reference blocks, a calculation means
`fer calculating a difference characteristic amount indicating a positive / negative sien of
`a difference between the correspondingpixel of the current block and the reference block
`for each pixel in the block and a difference feature amount calcwated for an adjacent
`reference block are compared. A comparator is provided with a comparison means for
`outputting 4 comparison result
`indicating a possibility that a shmilamty of an
`interpolation block obtained by interpolating a pixel between adjacent reference blocks
`is higher than a similarity of the adjacent reference blocks, and a determination maans
`for determining that the search of the 1 optimumblock is ended whenthe comparison
`yesult indicates a predeterminedpossibility.
`0624]
`
`Also, in the mation vector detection devices, the calculation means calculates a difference
`feature amount indicating only the positive and negative signs of the difference of the
`corresponding pixel for each pixel in the block, and the comparison means has the
`comparison means. The sign shawn with the aforementioned difference characteristic
`quantity outputs the numberof inharmonious pixels as the aforementioned comparison
`result, and the aforementioned judgment means may end seareh of
`the Ist
`aboveementioned optimum block, when the aforementioned comparison result is more
`than a predetermined threshold value.
`[Effect of the Invention!
`foo2s]
`Acverding to this configuration, the comparison reault indicating that the similarityof
`the interpolation bloek obtained between the adjacent reference blocks is higher by
`comparing the sign of the difference between the corresponding pixels of the adjacent
`reference blocks is calculated, so that the result can be obtained for each interpolation
`block.
`
`If the sien of the difference between the corresponding pixels of the adjacent reference
`
`

`

`block does not mateh, sinee there is a pixel value of the current block between the
`corresponding pixel values, the value of the interpolated pixel obtained by linearly
`interpolating the corresponding pixel is eclese to the value of the pixel of the eurrent
`block. "Phe comparison result, which is the number of such corresponding pixels,
`accurately indicates a possibility that the similarity of the interpolation block becomes
`higher.
`loo26]
`Further, since the partial search for the | optimal block is terminated based on the
`comparison result, it beecames possible to detect the preferred interpolationblockin the
`middle of the search path when the 1 optimal block is partially searched, and at the
`same time, the calculation amount requiredfor the partial search is reduced.
`tm addition, in the motion vector detecting device, the calculating means caleulates a
`difference feature amount indicating a quantization value of an absolute amount of a
`difference for each pixel in a block in addition te a positive and negative signof a
`difference of the corresponding pixel, and the comparing means calculates the difference
`characteristic
`amount. The
`sign
`shawn with
`the
`aforementioned difference
`characteristic quantity is inharmonious, and the murmber of pixels with which a
`quantized value corresponds is output as the alorernentioned conrparison result, and it
`may judge with the aforementioned judgment means ending search of
`the Ist
`above-mentioned optinrum block, when the aforementioned comparisonresult is more
`thana predetermined threshold vahue.
`10627]
`
`According to this configuration, the comparison resulf is a mamber of corresponding
`pixels in which the positive and negative signs af the difference do not match and the
`absolute amountof the difference matches with the resolution of the quantization. Since
`the value of the interpolated pixel obtained by lmearly interpolating the corresponding
`pixel becomes exactly the value of the pixel of the current block at the resolution of the
`quantization,
`if is possible to further aecurately indicate the possibility that the
`similarity of the interpolated block becomes higher.
`foogs|
`The motion vector detection method of the present Invention includes steps of setting a
`plurality of reference blocks in a reference image m stages,
`After a 1 optimumblock having the highest similarity with a current block in an image
`to be encoded is searchad out of the set reference blocks, one or more Interpolation
`blocks obtained by pixel interpolation between the reference blocks are added to the
`comparison object to obtain a 2 optimzum block having the highest aimilarity.
`
`A motion vector detection method for detecting a motion vector with an accuracy of an
`
`

`

`integer pixel or less based on a 2 optimum block, and for each of the set referenceblocks,
`a calculation step for calculating a difference characteristic amount indicating a positive
`/ negative sign of a difference between the corresponding pixel of the current block and
`the corresponding pixel of the reference block for each pixel in the block By comparing
`the calculated difference feature amounts with respect to the adjacent reference blocks,
`the calculated difference feature amounts are compared. Phe method includes an
`estimation step of estimating a higher degree of similarity of an interpolation block
`obtamed by interpolating pixel interpolation between adiacent reference blocks than a
`
`the search of the 1 optimal block is ended when the similarity is estimated to be
`positive.
`joo29]
`
`In addition, the calculation step maycalculate a difference feature amount indicating
`only the positive and negative signs of the difference of the corresponding pixel for each.
`pixel in the block, and the estimation step may be positively estimated whenthe number