CN1003413B - Improved system for coring an image-representing signal - Google Patents

Abstract

The present invention relates to a system for reducing noise in the display of an image for preventing obvious bad results by three measures. The three measures are carried out as follows: (1) obtaining the signals (I1 to IN) of sub frequency spectrums with the utilization of a frequency spectrum analyzer (200) without oscillation or aliasing, and with position conversion and an octave bandwidth; (2) utilizing dispersed coring devices (202-1 to 202-N) for coring the signals of one or more decomposed sub frequency spectrums; (3) utilizing one synthesizer (206) with one or more filters (204-1 to 204-N) without oscillation or aliasing for obtaining an output image representing signal (IR) from all of the signals of the sub frequency spectrums.

Description

The improvement system of coring image representation signal

The present invention relates to utilize the image processing system of coring technology,, for example be used for TV image signal to reduce the noise composition in the image representation signal.Particularly, the present invention is when reducing the noise composition of system and in unconspicuous alias or other the false spacing frequency composition introducing image representation signals.

Coring be a kind of everybody know, be used to reduce the technology of image representation signal noise composition.Coring is exactly that selectively the image representation signal that only allows absolute amplitude to surpass a certain set point passes through.Coring is a non-linear process, and it can be introduced in the image representation signal with the mutual spacing frequency composition of modulating by false harmonic wave originally.The power of these false spacing frequency compositions can increase with the increase of value.Therefore, the selection of coring value should be traded off between the two: value should be enough greatly to reduce the noise composition, and the while again can not be excessive, in order to avoid introduce too much false spacing frequency composition.

Observer for a displayed image, the influence that the noise composition of image representation signal is caused is depended on both: the relative size of the interval frequency spectrum of noise composition and signal composition in (1) displayed image, (2) are in the habit of experiencing people's vision system aspect the noise.

Known people's vision system is formed in the primary leading frequency of calculating luminousing image, and what lean on is that spacing frequency information is divided into many adjacent and overlapping separation band.Each bandwidth roughly is an octave, and the centre frequency of each frequency band and nearby frequency bands centre frequency roughly are the relations of twice.Studies show that have 7 frequency bands or " passage " to cross over the spacing frequency scope of the 0.5-6 circulation/degree of people's vision system approximately.Find further that again the spacing frequency processing that occurs in people's vision system is by sterically defined.Therefore the signal in each spacing frequency passage is that little subregion by image calculates.These subregions are overlapped, and the width of two circulations is roughly arranged on characteristic frequency.If, then can find as test pattern with-sinusoidal wave rastered picture, the contrast-sensitivity function of sinusoidal wave rastered picture tilts rapidly along with the raising of image spacing frequency.In other words, high spacing frequency needs high contrast (circulating 30/be about 20% when spending), and lower spacing frequency needs quite low contrast (circulating 3/be about 0.2% when spending).Know, people's vision system is observed the ability of the contrast variation of sinusoidal wave rastered picture (this image is on value), the situation that when lower spacing frequency, is better than higher spacing frequency, particularly, people's average speciality is, in order correctly to differentiate 75% of contrast variation, concerning the sinusoidal wave grating of one 3 circulation/degree, need to change contrast 12%, and concerning the grating of one 30 circulation/degree, then need change contrast 30%.

Function based on people's vision system, obviously, higher signal to noise ratio (S/N) is easy to cover noise (promptly concerning the observer, noise is negligible) in octave separation frequency band, and for higher spacing frequency octave, this to cover effect just more remarkable.This is because under higher spacing frequency, and people's vision system all descends aspect contrast sensitivity and changes in contrast sensitivity two to some extent.On the other hand, quite little high spacing frequency noise composition be added to one almost uniformly on the background (this background comprise direct current or very low tone every the video composition of frequency), be easy to by people's vision system observed.This point is very important, because the image of real world is on two orthogonal dimension, major part all has a frequency spectrum at interval, and this frequency spectrum contains a large amount of lower spacing frequency signal energies, has only the sub-fraction higher frequency signal energy.This will make any high spacing frequency noise become obvious especially.

Represent whole intervals frequency spectrum of signal if having only a coring device to be used for input imagery of coring, then selected value is too little probably and can not reduce tangible noise composition in the one or more octaves part of this interval frequency spectrum satisfactorily, and while other one or more octave parts in this interval frequency spectrum, it is too high that value can seem again, causes the false spacing frequency composition of excessive number to be introduced in the displayed image.

This problem can solve with following way, at first input imagery is represented that the frequency spectrum of signal decomposes, and makes it to be included in several adjacent sub-spectral bands.Distinguish these frequency bands of coring one by one then, make the corresponding suitable value of each frequency band.At last these frequency bands by coring are synthesized a single output image and represent signal, and signal draws displayed image thus.

No. 4442454 patent that the present invention April 10 delivered with reference to U.S. Powell1984.

Image processing system of the present invention can make input imagery represent that any noise composition that occurs in the signal spectrum is reduced in output signal spectrum.And avoid representing to introduce in the signal spectrum tangible alias or other false spacing frequency compositions at output image.Therefore the present invention need not take into account the artifact that merits attention itself that handle to some extent in order to reduce noise.

The brief description of illustration

Two figures shown in Figure 1 are represented " brick wall " mode filter characteristic and a filter characteristic that tilts gradually respectively.

Fig. 2 is the functional block diagram of idealized realization of the present invention.

Fig. 2 a is depicted as the distortion of the frequency spectrum synthesizer among Fig. 2.

Fig. 3 a illustrates a Burt Pyramid spectrum analyzer, is useful during its spectrum analyzer in specific implementation Fig. 2.

Fig. 3 b illustrates a Burt Pyramid synthesizer, is useful during its frequency spectrum synthesizer in specific implementation Fig. 2 a.

Figure shown in Figure 4 is the baseband envelope line chart of 7 input multiplication constant core rights functions, and these 7 coefficients have numerical value separately.

Fig. 5 is the calcspar of the more excellent implementation method of coring device shown in Figure 2, and this device is suitable for vision signal of coring, and this signal is to be used for limiting a two-dimentional TV image that is scanned.

The description of more excellent implementation method

Outstanding feature of the present invention is, each spectrum analyzer and synthesizer only comprise and have the filter of slant filtering characteristic gradually.Centre frequency f among Fig. 1 _cBe two cut-off frequency f ₁And f _hMean value.Concerning band pass filter, f ₁Greater than 0, concerning low pass filter, f ₁Equal 0.The problem of " brick wall " mode filter is, produces as out-of-band high-frequency impulse energy can cause vibration when excited by impact.This false signal can find out, because people's vision system is very responsive on the frequency background for low tone that high spacing frequency glitch is added to.

The filtering characteristic that the normalization of a band pass filter is tilted gradually is shown in figure 102, and its centre frequency is f _cBecause inclination is gradually, therefore there is not significantly to limit the bound frequency f of bandwidth ₁And f _h, but attenuated input signal is located to a certain degree (for example half-power point place) as f ₁And f _hBandwidth is exactly specified bound frequency f like this _hWith f ₁Poor.Yet shown in the shadow region 104 of Fig. 1, use the spectrum analyzer of this filter, under a given sub-spectral bandwidth, will have the sub-fraction energy that contiguous sub-spectral bandwidth is caused interference.This situation can cause false spacing frequency in the image processing system that uses sampling and sub-sampled signal.And image processing system of the present invention reduces any alias effect with as described below.

The band pass filter that tilts gradually for shown in the figure 102 of Fig. 1 tilts to usually occur in centre frequency f _cHigh frequency limit and low frequency limit.Yet in fact have only the high frequency limit of centre frequency to tilt.The accurate shape of any specific characteristic of slant filtering gradually is a design problem.Image processing system of the present invention is used gradually, and the design criterion of slant filtering device will go through below.

Shown in Figure 2 is the functional block diagram of ideal scheme of the present invention.Among the figure, dead-beat, the conversion of no alias location, the spectrum analyzer 200 of octave bandwidth has an image representation signal 1 as input.In fact, input signal I can be a continuous analog signal, a sampled analog signal (for example CCD imager and signal converter are used) or sampled digital signal (for example deriving from analog to digital converter).In fact the image processing type that will discuss is almost always represented signal at sampling image, this processing is to be undertaken by the non real-time mode by the spectrum analyzer that utilizes digital computer, or undertaken by real-time or non real-time mode by the spectrum analyzer that utilizes physical hardware.Therefore, for convenience of explanation, input signal I is a sampled signal but not continuous signal.

As shown in Figure 2, image representation signal I is limited in the one dimension that is expressed image by a frequency spectrum at least, and the scope of this frequency spectrum is from highest frequency f _mTo 0.For not comprised, signal I is higher than f _mSpacing frequency, suppose this signal by a prefilter, for ease of explanation, putative signal I is a temporal vision signal, this vision signal is from the two-dimentional TV image of a plain scan (although this and nonessential).In a word, analyzer 200 resolves into N(N for greater than 1 integer with the spacing frequency of input signal I) the logical sub-frequency spectrum output signal L in individual phase adjacent band ₀L _-1With a sub-spectrum signal GN of remnants, L ₀L is by the spacing frequency sequence arrangement that falls progressively, from f _mBeginning is up to bandwidth zero, that each signal is composed every a frequency multiplication successively.GN comprises all spacing frequencies of the frequency spectrum of input signal I, and these frequencies are lower than all frequencies (the logical frequency spectrum of band that the logical sub-frequency spectrum of this band is minimum spacing frequency) that are included in the logical sub-frequency spectrum of (N-1) individual band.Particularly as shown in Figure 2, octave 1 has f _m/ 2 nominal bandwidth and 3f _m/ 4 centre frequency, octave 2 has f _m/ 4 bandwidth and 3f _m/ 8 centre frequency, the rest may be inferred.

Each coring device 202-1-202-N is corresponding to sub-frequency spectrum output signal L ₀L _-1With among the GN one.Coring device 202-1 ... each signal L of 202-N output ₀L _-1Add to corresponding each dead-beat of frequency spectrum synthesizer 206, no aliasing filter respectively with GN.Frequency spectrum synthesizer 206 also comprises an adder 208, is used for filter 204-1 ... the output addition of 204-N is represented signal I with the output image that obtains a reformulation.

The image spacing frequency spectrum of 200 couples of input signal I of spectrum analyzer carries out linear transformation.Therefore, provide at spectrum analyzer 200 under the situation of location conversion of dead-beat, no alias, baseband slot frequency few in number will appear on each output of spectrum analyzer, appear at also few in the image spacing frequency spectrum of signal I.Therefore, the false spacing frequency composition of being introduced by spectrum analyzer 200 and not obvious.Yet, be operated in the coring device 202-1 of nonlinear state ... 202-N must introduce each output signal L with false spacing frequency composition ₀L _-1In the middle of GN.These false compositions are at interval made by the harmonic wave composition of sub-spectrum intervals frequency and mutual key that part forms, and this a little spectrum intervals frequency inputs to coring device 202-1 respectively ... 202-N.All harmonic wave compositions of any spacing frequency in the sub-frequency spectrum of octave bandwidth have the spacing frequency that is positioned on the sub-frequency spectrum of this octave bandwidth.

If the output of the coring device that sub-frequency spectrum input is operated to octave bandwidth is added on the filter with " brick wall " characteristic (seeing the figure 100 of Fig. 1), then the harmonic wave that is produced by coring device and all false spacing frequencies of mutual modulation composition filter filtered device.

Yet, for above-mentioned reasons.This filter with " brick wall " characteristic is easy to introduce the false spacing frequency vibration composition that causes excited by impact.For fear of introducing this composition, this band pass filter should have the filtering characteristic (shown in the figure among Fig. 1 102) of inclination gradually and the nominal bandwidth of an octave.At this moment the spacing frequency composition that a spot of harmonic wave and mutual modulation cause will can all not suppressed, and this is owing to exist the outer part of band (shown in the dash area among Fig. 1 104) of filtering characteristic.Yet, will go through below, the caused false spacing frequency composition of the characteristic of Qing Xieing can rely on special Design of Filter to be eliminated (promptly can not discovering basically) on displayed image gradually.

Each filter 204-1 of spectrum analyzer (206) ... 204(N-1) can be band pass filter or low pass filter.If band pass filter, then each filter all has a centre frequency and the bandwidth corresponding to octave spectrum.If low pass filter, then each filter all has one from zero bandwidth to specified upper cut off frequency, and this bandwidth is identical with the bandwidth of the respective band pass filters with same octave spectrum.

Filter 204-N with remaining sub-frequency spectrum is a low pass filter, the lower limiting frequency of the sub-frequency spectrum of its specified upper cut off frequency as many as (n-1) individual octave.

If low pass (but not band is logical) filter is as octave filter 204-1 ... 204-(N-1), then the false spacing frequency composition of low strap difference frequency (beat frequency) of nucleating course can't suppress.Yet the mutual modulation signal of this beat frequency will be low level, thereby appear in the displayed image and just to be difficult for being discovered by people's vision system.This be since the real world image in the higher level signal composition be positioned at the image spacing frequency spectrum low tone every the frequency part, thereby form a kind of masking effect.Moreover in real system, low pass filter is than the suitable dead-beat of the easier realization of band pass filter, the slope characteristic gradually of no alias.

In Fig. 2,, also nonessential for purposes of the invention like this although the sub-frequency spectrum output signal of each of analyzer 200 all has an independent coring device.All required is exactly that at least one sub-frequency spectrum output signal has independent coring device.Yet, if coring device 202-2 ... among the 202-N any one has that this is the highest interval frequency spectrum by being lower than octave 1() the frequency spectrum formed of spacing frequency, then this device must have the respective filter of a frequency spectrum synthesizer 206, so that remove at least in the frequency band because the false spacing frequency composition that is caused is handled in nonlinear coring.But under the situation of octave 1 sub-frequency spectrum, often occupy the filter 204-1 of frequency spectrum synthesizer 206.This is because most of image shows that the spacing frequency of being differentiated can not be higher than the highest spacing frequency f of octave 1 sub-frequency spectrum _mBecause among the synthesized output signal I, the arbitrary false spacing frequency composition of a last frequency band all can not be differentiated, thereby concerning this special circumstances, do not need the composition of filtering frequency band on this in the image demonstration.

Whether need to comprise the filter of an independent dead-beat, no alias in the frequency spectrum synthesizer 206, make it to link, can select arbitrarily with the sub-frequency spectrum output of any one omission coring.Yet adder 208 is with whole N band pass filter and the signal plus frequency spectrum of remnants.And still save filtering no matter always in the signal that any one sub-frequency spectrum of spectrum analyzer 200 is exported, save coring.

Now referring to Fig. 2 a, the figure shows out a frequency spectrum synthesizer 206a who changes picture, the frequency spectrum synthesizer 206 of its available Fig. 2 is replaced, and synthesizer 206a has used a plurality of low pass filter 210-2 ... 210-N and half adder 212-1 ... 212-(N-1).The specified upper cut off frequency of low pass filter 210-2 equals image representation signal I(, and this is the upper cut off frequency of frequency multiplication 2 sub-frequency spectrums) the largest interval frequency f of image spacing frequency spectrum _mPartly.Similarly, low pass filter 210-3 ... the specified upper cut off frequency of each equals the upper cut off frequency of the octave spectrum that links with this filter 210-(N-1).The specified upper cut off frequency of low pass filter 210-N equals the lower limiting frequency of (N-1) individual octave spectrum.

In Fig. 2 a, filter and the half adder arranged by reverse order close mutually by the series connection form, and the result is that (the residue signal G ' of coring is N) successively one by one by each series connection low pass filter filtering for minimum interval spectrum signal.Shown in Fig. 2 a, the next signal L ' of minimum interval spectrum signal _-1With the output signal of filter 210-N by half adder 212-(N-1) carry out addition, then successively by each low pass filter filtering.Similarly, the sub-spectrum signal L ' of the spacing frequency octave of coring ( _-2) each signal among the L ' 2 is all successively by all filter filterings of frequency spectrum synthesizer 206a, shown in Fig. 2 a, last, the output of filter 210-2 is through the sub-spectrum signal L of the highest spacing frequency octave of half adder 212-1 and coring ₀Addition is represented signal I to obtain an output image that reconstitutes.In Fig. 2 a, suppose that visual demonstration can not differentiate input imagery frequency spectrum largest interval frequency f _mOn spacing frequency, therefore do not need to provide a low pass filter for the output of half adder 212-1.

According to the discussion of Fig. 2 and 2a as can be known, essence of the present invention is that spectrum analyzer 200 is no longer exported sub-spectrum signal except that residue signal, and this will have a nominal bandwidth less than an octave.Yet in the logical sub-frequency spectrum letter of the band that principle of the present invention is used for analyzing, each all has the spacing frequency bandwidth less than an octave.

Burt Pyramid spectrum analyzer and Burt Pyramid frequency spectrum synthesizer are suitable to especially spectrum analyzer 200 and frequency spectrum synthesizer 206a among the present invention, and reason has two at least.The first, the characteristic that this Burt Pyramid allows filter to have to tilt gradually but not " brick wall " characteristic, thus be suitable for as frequency analyzer and frequency spectrum synthesizer.The second, in optimal case of the present invention, Burt Pyramid spectrum analyzer produces the logical sub-frequency spectrum output signal of specified octave bandwidth band and the sub-frequency spectrum output signal of remnants.

Burt Pyramid analyzer is analyzed the input signal Go of sampling.In the following explanation of relevant Burt Pyramid, suppose that Go(is equivalent to the image representation input signal I among Fig. 2) form be the conventional vision signal (for example TSC-system vision signal) of the determined interval of continuous sweep bidimensional TV image frequency spectrum, this video letter at first through filtering in advance to remove any the highest given spacing frequency f that is higher than _mThe spacing frequency composition, then at least two times of f _mSample frequency under be sampled.

Burt Pyramid analyzer is shown in the function diagram of Fig. 3 a in real time.Fig. 3 a shows that analyzer contains one by identical substantially sampled signal transmitting stage 300-1,300-2 ... the streamline that 300-N formed.Each level work is by digital dock CL ₁, CL ₂On the determined sample frequency of CL.Be added to clock frequency on any one-level and all be lower than the clock frequency value of its prime.For purposes of the invention, 300-2 ... the clock frequency value of each of 300-N grade all is half of its previous stage clock frequency value.

Shown in Fig. 3 a, the 300-1 level contains convolution filtering device and damping characteristic device 302, deferred mount 304, substracting unit 306 and expansion and interpolation filtering device 308.One with clock CL ₁The input traffic of the digital sample values Go that samples of frequency be added into by convolution filtering device and attenuating device 302, to obtain one with clock CL ₂The digital sample values G that samples of frequency ₁Output stream.The convolution filtering utensil has lowpass function, can be with G ₁The visual fibrillar center of each of representative frequency is reduced to half of centre frequency of the respective dimension of Go representative.Simultaneously, attenuating device reduces half with the sampling density of each dimension.

Each digital sample values of Go is added on the substracting unit 306 by deferred mount 304 as first input.Simultaneously, G ₁The sampled value of reduction density add to expansion and interpolation filter 308, it has improved G ₁The sampling density of sampled value, and turn back to the sampled value of Go.Then, interpolation G ₁Density after the expansion of sampled value adds to substracting unit 306 as second input.The existence of deferred mount 304 has guaranteed G corresponding mutually on interval location ₀And G ₁Each sampled value is added in first and second inputs of substracting unit 306, and coincide mutually in time.Continuous sampling value L from substracting unit 306 ₀Output stream define scanned picture each the dimension in the highest spacing frequency.

300-2 ... 300-N structure at different levels is identical with the 300-1 level basically.Yet, 300-2 ... during 300-N was at different levels, the higher level of each sequence number was handled than the signal of low tone every frequency, and this signals sampling density is lower than the sampling density of its previous stage.Particularly, continuous sampling value L ₁Output stream represented in each image dimension the next octave of the highest octave of frequency at interval, therefore shown in Fig. 3 a, the signal that decomposes through Burt Pyramid is by each octave sampled data stream L ₀L _-1(derive from 300-1 respectively ... the substracting unit that 300-N is at different levels) and low frequency residue signal G(derive from the convolution filtering device of 300-N level and the output of attenuating device) formed.Fig. 3 b shown one with Fig. 2 b in the corresponding Burt Pyramid of synthesizer 206a synthesizer, be used for the output signal I that obtains to reformulate.This is by utilizing coring sampled data stream L ' ₀L ' _-1The time-delay mechanism 310-0 that operates ... 310-(N-1), and expansion and interpolation filter 312-1 ... 312-N and adder 314-0 ... 314-(N-1) finish.Above-mentioned as institute, serve as reasons two times of represented each image spacing dimension of expansion and interpolation filter 312-N of the sampling density that the remaining sampled data stream GN of least density has are then via adder 314-(N-1) and the sampled data stream L of a demonstration _-1Addition.This process is carried out repeatedly by continuous synthetic level, has drawn the output signal L that reconfigures, and the bidimensional image is shown due to high sampling density.

Used each convolution filtering device and the interpolation filter of Burt Pyramid all has low pass filter, and they must satisfy following two conditions.The first, each this filter uses a symmetrical core weighting function that comprises three multiplication constants at least.The second, each multiplication constant of core rights function must play a part identical, in other words, must provide identical total power to each node that is positioned at next higher level to each node of a preset level.To the requirement of the filtering characteristic aspect that actual dead-beat, no alias, location change is provided, as of the presently claimed invention, to the Burt Pyramid analyzer that uses for the present invention and the convolution and the used core rights function of interpolation low-pass filter of synthesizer additional restrictions has been proposed again.Just consider these conditions now.

Because the operand of each grade of the Burt Pyramid analyzer shown in Fig. 3 a all is the input signal of sampling, therefore will consider that sort signal not only comprises the Base Band part, and comprise the repeating part that each contains a lower sideband and upper sideband.In addition, in order to prevent alias, sample frequency should be at least and adds to the Burt Pyramid analyzer sampled input signal Base Band at different levels twice of the highest frequency of frequency spectrum at interval.

Utilize mathematical analysis to express, concerning Base Band, the 300-1 of Burt Pyramid spectrum analyzer ... each grade during 300-N is at different levels all is similar to the device of a dead-beat, no alias, location conversion, when following relation is set up, can be used to obtain the signal L that decomposes through Burt Pyramid ₀L _-1In have of corresponding octave bandwidth.These relations are:

1. within spacing frequency scope 0≤f≤fh/4 (fh is the specified upper cut off frequency of the sub-spectrum signal of each octave in the formula), the product of every grade the convolution filtering device and the normalization transmission characteristic of interpolation filter is 1.Particularly, every grade convolution filtering device and any one in the interpolation filter, the transmission characteristic in this scope all are 1.

2. within spacing frequency scope fh≤f≤3fh/4, the product of every grade the convolution filtering device and the normalization transmission characteristic of interpolation filter is decayed gradually.Particularly, every grade convolution filtering device and any one in the interpolation filter all have the characteristic of decay gradually in this scope.

3. within spacing frequency scope 3fh/4≤f≤fh, the product of every grade the convolution filtering device and the normalization transmission characteristic of interpolation filter is zero.Particularly, every grade convolution filtering device and any one in the interpolation filter, the transmission characteristic in this scope all are zero.

Have only the convolution filtering device of the 300-N level of Burt Pyramid analyzer to be used to produce remaining sub-spectrum signal GN.Therefore, concerning the 300-N level, must make the convolution filtering device really meet above-mentioned condition 1,2 and 3.In addition, each interpolation filter of the synthesizer of Burt Pyramid shown in Fig. 3 b also should meet these three conditions (wherein, concerning synthesizer, fh is the adder output signal upper cut off frequency, and this adder is immediately following after the interpolation filter shown in Fig. 3 b).

The actual convolution and the design of interpolation filter not only must meet above-mentioned three conditions, and must meet the restrictive condition of above-mentioned symmetry and even distribution core rights function.This Burt Pyramid sampled signal convolution of core rights function or the interpolation filter of utilizing can not satisfy above-mentioned condition 1 and 3.Just equal two times the situation (this is the minimum sample frequency that prevents that alias is required) of the specified upper cut off frequency of an octave for sample frequency, can design a sampled signal Burt Pyramid convolution or an interpolation filter with core rights function of 5 multiplication constants, it can satisfy above-mentioned condition 1 or 2, but can not the both satisfy.In order to design the Burt Pyramid convolution or an interpolation filter that can satisfy all above-mentioned three conditions, need have the sampled signal filter of the core rights function of 7 multiplication constants at least.Particularly, just be the sample frequency of the twice of the specified upper cut off frequency fh of filter input signal frequency spectrum for having in no alias value, when the value of variable P for-when 0.052(sees that Fig. 4 a), Base Band filtering characteristic shown in Figure 4 (it meets all above-mentioned three conditions) is by 7 shown in Fig. 4 a multiplication functions, symmetry and equally distributed filter kernel weight function limits.

For situation about being discussed (being that sample frequency is the situation with specified minimum no alias), the core rights function of 7 multiplication constants can provide real location conversion, because this function is only operated quite local image region.(it is normally enough little for the core rights function of 9 multiplication constants, provide real location conversion with the no alias sample frequency of specified minimum for it as before) be desirable, because it is next compared with the core rights function of 7 multiplication constants at the vpg connection of the sloping portion gradually that limits the Base Band filtering characteristic, can provide good tuning capability.

Yet, produce the no alias value of specified minimum in order to make sample frequency, when the quantity of the multiplication constant of filter kernel weight function greater than 9 the time, filter transform tends to not have locate mode day by day, and this is undesirable.On the other hand, if filter signal is over-sampling (being the twice that sample frequency substantially exceeds the specified upper cut off frequency of filter input signal frequency spectrum), then the quantity of multiplication constant must increase in the core rights function, so that the same filter characteristic with same location degree to be provided.For example, if the sample frequency shown in Fig. 4 and the 4a is 4 times of f ' h, then the core rights function that tool is identical with the envelope of core rights function shown in Fig. 4 a will be made of (in other words, an interpolate value will be inserted between any a pair of adjacent multiplication constant shown in Fig. 4 a) 13 to 15 multiplication constants.

Because coring is a non-linear process, and introduce artifact, the coring outside therefore need not wishing is to remove the noise that appears in the coring device input signal.In addition, coring value should remain on to be enough to remove on the minimum level of this noise.Fig. 5 is coring device 202-1 ... the calcspar of the desirable scheme of 202-N, these coring devices are used to represent to scan standard (for example NTSC) vision signal of two-dimentional TV image.Known in TV, this TV signal comprises continuous interlaced field, and each scanning field partly is made up of effective scanning field part and vertical blanking.Transitive graph image information during effective field is at black-out intervals transitive graph image information not.The noise that black-out intervals occurred of each scanning field can be used as the tolerance of the noise of next valid period.

As shown in Figure 5, the value of coring device 202-K(K is 1 ... N) contain switch 500, only the vertical blanking part at each just applies control signal with this switch connection.Therefore in the whole valid period of each, switch 500 is all opened a way.The L that links with coring device 202-K _-1This sub-frequency spectrum input signal is added to switch 500 and nonlinear amplifier 502.Therefore, has only L _-1The switch 500 that is closed of noise section be sent to rectifier 504(because at vertical blanking period, signal only comprises noise).Noise section through rectifier 504 rectifications is sent to integrator 506, and its time constant is t _kt _kEnough big, with extend by rectification at the noise contribution that vertical blanking period took place, make it to cover fully the live part of next.Therefore, integrator 506 produces a live part use with direct current signal of adjustable value for a field.The noise level of the blanking part of this direct current signal and previous field is proportional.The signal that this is adjustable adds to nonlinear amplifier 502 as a control signal.In addition, for the total signal gain for each coring device compensation different value, the gain of the nonlinear amplifier of each coring device can be regulated by gain controlling respectively.Particularly, on duty when big, the coring output signal L ' of nonlinear amplifier 502 _K-1Relative power just can be less than input signal L _K-1Relative power.Therefore, the purpose of gain-adjusted compensation is, makes sub-spectrum signal (the output L ' of nonlinear amplifier 500 for example of each independent coring _K-1) offer the power and corresponding L of synthesizer output signal I _K-1The power that sub-spectrum signal offers the input signal I of spectrum analyzer keeps equating.

The work of nonlinear amplifier 502 is amplification input signal L only _K-1In, absolute level surpasses that part of existing adjustable value.Therefore, even when the absolute level of input signal has surpassed existing adjustable value, the part of also having only the quilt of the input signal of exceedance to prune just can be passed through nonlinear amplifier 502, and to coring device L _K-1Output signal power is provided.Another kind method is that the absolute level of input signal and adjustable value are compared.If input signal absolute value exceedance, then input signal will be all by arriving L ' _K-1Output, otherwise will can pass through without any signal.The advantage of this method is that quite a large amount of input signal powers are retained in the output signal power.Yet in the past, a shortcoming of this method is that this method is easy to produce a high alias artifact that is called " spark " in displayed image, and this image is got by this coring device.Yet an image processing system that combines with principle of the present invention can suppress " spark ", thereby makes this method more practical.

Claims (13)

1, a kind ofly is used to handle the image processing system that an input imagery is represented signal (I), wherein said image representation signal has defined many spacing frequencies on the one dimension of represented at least image spectrum, these spacing frequencies are in the scope of highest frequency fm to O, wherein said image representation signal has been carried out spectral decomposition, begins to resolve in proper order by the spacing frequency that falls progressively the logical sub-spectrum signal (L in phase adjacent band of one group of one or more separation from fm ₀, L ₁...), each sub-spectrum signal has a specified bandwidth, and its width is not more than an octave in the described scope; And resolving into a sub-spectrum signal of remnants (GN), the sub-frequency signal of these remnants has comprised all that spacing frequency of the described image representation signal spectrum under the frequency spectrum in being included in the logical spectrum signal of minimum spacing frequency band; And wherein said system comprises: the device (202-1 that is coupled to the image representation signal of described spectral decomposition ... 202N), it is used at least one described sub-spectrum signal of coring; With a frequency spectrum synthesizer (206,206a), it is by connecting receiving described sub-spectrum signal, thereby and comprised by connecting to receive the adder that described sub-spectrum signal forms an image representation output signal (I); It is characterized in that, for fear of false or other parasitic spacing frequency compositions of in the frequency spectrum of described output signal, introducing significant quantity, described frequency spectrum synthesizer comprises: have essentially no vibration, the filter apparatus (204-1 of no alias translation function characteristic ... 204N; 210-2 ... 210N), this filter apparatus be coupled at least each the frequency of coring signal be than the low sub-frequency spectrum of the sub-frequency spectrum output signal of highest frequency; And described adder (208,212-1 ... 212(N-1)) with all described sub-spectrum signal additions, this a little spectrum signal both comprised those by coring with those also not by sub-spectrum signal coring and/or filtering, also comprise any both not by the also not filtered sub-spectrum signal of coring.

2, the system as claimed in claim 1 is characterized in that the signal that the highest described spacing frequency octave spectrum decomposes is directly to be coupled to described adder (208 from described coring device (202-N); 212-N).

3, the system as claimed in claim 1 is characterized in that each described synthesizer filter apparatus comprises a low pass filter (204-N; 210-N), this filter equal sub-frequency spectrum on the nominal cut-off frequency place of spacing frequency the frequency characteristic that descends is gradually arranged, this filter apparatus promptly has been coupled to this separately frequency spectrum.

4, system as claimed in claim 3 is characterized in that: described synthesizer comprise at least two by an adder (212-2) cascade described low pass filters (210-2 for example, 210-3), this adder is included in the described adder; First described filter (210-3) separately with described sub-spectrum signal in one relatively the signal of low frequency (L ' _-1) link, and second described filter (210-2) separately with described sub-spectrum signal in a more high-frequency signal (L ') link; A more low-frequency signal is as importing (through 212 in the described sub-spectrum signal _-1) add to described first low pass filter; One output of described first low pass filter adds to described adder (212-2) as first input; A more high-frequency signal adds to described adder as second input in the described sub-spectrum signal; An output of described adder adds to described second low pass filter (210-2) as an input, and a more low-frequency signal is by the first and second low pass filter filtering of described two cascades in the described so sub-frequency signal.

5, image processing system as claimed in claim 1 is characterized in that: be the vision signal of the image that has been scanned at described at least one dimension of an expression one by the described image representation signal of spectral decomposition; Described vision signal does not comprise the instantaneous frequency corresponding to an image spacing frequency that is higher than fm, and described vision signal is sampled by the instantaneous sample frequency corresponding at least 2 times of fm; The logical sub-spectrum signal group of described band comprises N the logical sub-spectrum signal of band, and N is the integer greater than 1 here; Described synthesizer is a Burt Pyramid synthesizer, and it comprises one group of N according to the order of sequence expansion-interpolated filter device (312-1 ... 312N) and adder (340-0 ... 314(N-1)).

6, image processing system as claimed in claim 1, its feature also is by a location transform analysis device that is coupled to the input of described image processing system described input imagery to be represented that signal carries out spectral decomposition, described analyzer comprise have essentially no vibration, the filter apparatus of no alias translation function characteristic.

7, as claim 1 or 5 described image processing systems, its feature also is: described input imagery represents that signal is the vision signal of expression one image, this image is scanned at described at least one dimension, and described vision signal is sampled by the instantaneous sampling frequency corresponding at least 2 times of fm; Described spectrum analyzer is a Burt Pyramid spectrum analyzer, and it comprises that one-level (300-1) is in order to obtain high-frequency signal (L the described sub-frequency spectrum output signal from it ₀), described one-level comprises:

(1) convolution filtering attenuating device (302), it is in response to the described vision signal (G that is sampled ₀), so that draw the first filtering output signal (G from it by the sample frequency of 1/2nd described vision signal ₁);

(2) expansion-interpolated filter device (308), the described first filtering output signal is to add to this filter apparatus as input, thus it draws the second filtering output signal by the sample frequency identical with described vision signal, and

(3) substracting unit (304) deducts the level value of each sampling of the described second filtering output signal in order to the level value from the corresponding sampling of described vision signal, thereby draws the high-frequency signal (L in the described sub-frequency spectrum output signal ₀), as the output signal of substracting unit.

8, image processing system as claimed in claim 7 is characterized in that: described Burt Pyramid analyzer comprises N level (300-1 ... 300N), N is a integer greater than 1 here.

9, as claim 1,5 or 7 described image processing systems, it is characterized in that described input imagery represents that signal is a vision signal, this vision signal is illustrated in the two-dimensional image that described two dimension all has been scanned.

10, image processing system as claimed in claim 9, wherein said vision signal are television raster scan video signals of being made up of the scanning field that recurs, and each described scanning field comprises the back with the blanking part that efficient image part is arranged; It is characterized in that each described coring device comprises: by switching device (500) and time constant device (506) form first the device, in order to obtain an adjustable threshold control signal (* _k), the value of this control signal is the direct function of the noise level of sub-frequency spectrum output signal during the efficient image part of each scanning field, this sub-frequency spectrum output signal is added to this coring device separately as input signal during the blanking part of this scanning field; Also comprise second device (502), this device is by the control of described threshold values control signal, if in order to only at the input signal (L of this coring device during each scanning field _K-1) this scanning field of surpassing of level during the value of described threshold value control signal the time, from this coring device draw an output signal (L ' _K-1).

Images