CN100466740C - A method to quantify difference for video code processing and its device - Google Patents

Abstract

The provided nonlinear differential quantizer for video coding system comprises: a first differential quantization unit with a quantizing subcell to convert every difference between near pixel values of video information into relative discrete step length to be scaled for full size of pixel amplitude, an overflow checkup subcell to check and correct the upper and/or overflow according to pixel amplitude. This invention can generate nonlinear feature coincident with human visual reaction system, and holds all HF components for well clearness and resolution.

Description

Be used for method to quantify difference and device thereof that video coding is handled

Technical field

Present invention relates in general to video coding and compression field, the method to quantify difference and the device thereof that use in handling more specifically to video coding.

Background technology

Existing video coding is realized many algorithms based on frequency domain, and for example those use the encoder algorithm of cosine transform in frequency domain.But there are some defectives in these encoder algorithms.For example, the semiconductor of real-time frequency domain encoder is realized, needs to surpass 800 ten thousand transistor usually.The Frequency Domain Coding device uses quantized frequency component to reduce coding information quantity.This may cause losing of some high fdrequency component, thereby reduces the definition and the resolution of video.VICOMP (video coding and compressibility and device) be one based on the digital video coding scheme under the demand of high-quality, real-time, high efficiency video coding (high compression rate).Its utilization is based on time domain but not the algorithm of frequency domain.Compare with above-mentioned Frequency Domain Coding device, VICOMP is for realizing that being less than 700,000 transistorized real-time encoders provides possibility.Simultaneously, compare with the Frequency Domain Coding device, VICOMP can provide same or better video quality and outstanding overall compression rate.The simplification of VICOMP on logic and complexity compared with Frequency Domain Coding, equally also makes the software of video coding system realize more effective.

For example, mpeg decoder all is based on the frequency domain information that uses discrete cosine transform (DCT) algorithm, and DCT is used for time-domain information is transformed into frequency domain.VICOMP is a kind of time domain coding device, is unnecessary with information from the step that time domain is transformed into frequency domain therefore.Save this step significant benefits is provided, because this shift step needs a large amount of calculating very at a high speed.

For example, for the block of pixels with 8 X 8 is transformed into frequency domain from time domain, needs to use the DCT algorithm to carry out 64 X and multiply each other for 64=4096 time.When tangible in the video coding, distribute to the time quantum that multiplies each other for 4096 times less than 5 microseconds.So single multiplier must be to be higher than the frequencies operations of 800MHz, perhaps 8 multipliers must be simultaneously to be higher than the frequencies operations of 100MHz, so that carry out this switch process in real time with each.

And the VICOMP decoder can utilize the clock frequency of 2X pixel rate to operate, and this clock is 27MHz typically, and by comparison, the DCT encoder is operated under the clock frequency of 100MHz.The difference of this clock frequency causes the less power consumption of semiconductor decoder.

Be based on a scheme of redundancy concept as above-mentioned VICOMP efficient coding scheme.The purpose of any one encoding scheme all is to find and portray redundant elements, generates an effective sets of signals then the nonredundancy signal is encoded.If in information, there is not redundant elements, also to compress with regard to having no chance. the redundancy of digital video signal results from the following aspects.

1. to the over-sampling of low frequency component---in the sampled digital vision signal, through regular meeting because sampling causes bulk redundancy.The sampling redundancy that part is such be because have to for the high-frequency response that guarantees sufficient resolution and digital sample to carry out video signal sampling in the frequency of 2 times of the maximum operating frequencies of analog video signal.But because contain the low-frequency component that accounts for main quantity in the real video signal, 2 times sample frequency can cause the over-sampling to low frequency component.The redundancy that therefore, in sampled signal, can have considerable quantity.

2. to the over-sampling of bit resolution---the redundant over-sampling that also can result from resolution (figure place) to each sample.Such as, if being digitized as 8 bit resolutions, each pixel multiply by 3 color coordinatess (as YUV) sampling resolution of 24 of each pixels just.24 that is to say have enough resolution to represent 1,670 ten thousand kinds of possible colors here.Sample for the pixel resolution with 640H X 480V in each independent frame of video, can produce 307,200 pixels (sample).Even can also only need 19 and represent 307,200 all colors with 307,200 all in frame pixels as different colors.Because picture is to form with the object of continuous relatively color, so in fact be less than 307,200 kinds of colors far away in a frame of video, the information that therefore also just need be less than 19 is illustrated in colors all in the frame.Furtherly, because the limited dynamic range of imageing sensor, 24/sample provides than the common dynamic range that is applied in the imageing sensor in the video camera (9) and has wanted the high dynamic range that manys (144 decibels) for 54 decibels.

3. interframe redundancy---the another kind of redundancy in the sport video is exactly that interframe repeats. and more a series of video frame images, we can find that the difference of common interframe is in very little scope. not all elements in video all changes along with the variation of frame.A lot of elements move very for a short time.For the rest image of successive frame, get rid of the factor of random noise, have 100% redundancy.Though the interframe redundancy can estimate that generally it generally is decided by the actual video situation, simultaneously the interframe redundancy range can from 0 to 100%.

4. noise filtering---because the limited dynamic range of video image sensors includes a large number of white noise in the video. this white noise is a completely random, that is to say that it has reduced redundancy and also reduced compression ratio.VICOMP is used for improving the senior filter of signal to noise ratio, since improve redundant and improve compression ratio.

5. redundant between line---each bar is given in video horizontal line exist a lot of with on element just the same or about the same in or next bar.Can by utilize specific algorithm find those with adjacent lines on different element and be its coding.

6. frequency of occurrences redundancy---some video elementary can be had the higher frequency of occurrences than other elements by expection.Can use the long code of variable bit is various element codings.For more frequent element occurring, those distribute minimum figure place.By using the long code of variable bit, the data bit sum that coding needs has reduced.

The above 6 kinds of elements that find, portray and encode can make information not lose or lose less.Except redundancy listed above, in digital video, still there is unnecessary information.That is to say the information that those do not have contribution to the quality and the key data of video.In order to improve compression ratio, need to eliminate or reduce these unnecessary information.

Summary of the invention

An important step that reduces inessential information coding is non-linear difference quantizer among the VICOMP of the present invention.When between adjacent two elements (in color and/or brightness) when having little difference, yet keep the accuracy of these little differences to be very important., the difference when between these adjacent elements is very big, and what keep these differences accurately is unessential.This is true, because the human vision reaction system is non-linear.As the human eye perception, adjacent little difference is unusual part and parcel for the texture of difference visual pattern.The absolute amplitude of adjacent big difference is not that an amount in the visual pattern is wanted factor.In order to utilize the non-linear of human vision reaction system, VICOMP uses non-linear differential quantizer.

VICOMP of the present invention uses the quantitative information based on the amplitude yardstick, uses unique nonlinear quantizer, produce with the human vision reaction system match non-linear.VICOMP has kept all high-frequency compositions, has therefore also kept definition and the resolution same with incoming video signal.Because the information that causes of nonlinear quantizer reduces and is not by the naked eyes perception.VICOMP of the present invention uses the over-sampling of its algorithm utilization for low frequency signal.

More generally, the invention provides a kind of non-linear differential quantizer that is used for video coding system, comprise the first difference quantifying unit, it comprises: quantize subelement, be used for each residual quantity of enter video information adjacent pixels input value is converted into a plurality of discrete steps, described discrete steps is calibrated full-scale with corresponding to pixel input range value suitably; The overflow check subelement is used for checking and proofread and correct up and down based on pixel input range value full-scale and overflows.

The present invention also provides a kind of non-linear method to quantify difference that is used in the video coding system, comprise step: each residual quantity of adjacent pixels input value in the enter video information is converted into a plurality of discrete steps, and described discrete steps is calibrated full-scale with corresponding to pixel input range value suitably; Check and proofread and correct up and down based on pixel input range value full-scale and overflow.

Description of drawings

By in conjunction with the detailed description of this accompanying drawing to the embodiment of video coding filtering method of the present invention and device thereof, purpose of the present invention, feature and benefit will be clearer.In the drawings:

Fig. 1 is the schematic block diagram that the typical case of VICOMP encoder of the present invention realizes;

Fig. 2 illustrates the signal logical flow chart that little difference 3 pixel groups residual quantity of the present invention quantize and make up;

Fig. 3 A-3B is the signal logical flow chart that shows non-linear difference quantization encoding of the present invention;

Fig. 4 is the schematic block diagram that shows the non-linear differential quantizer that originally is used to realize non-linear difference quantization encoding of the present invention.

Embodiment

For the ease of understanding concrete technical scheme of the present invention, at first related concrete technological concept is defined.

Definition:

Quantization---information translation is become the discrete form different with original form.Can be become 2,4,6 by quantization with following algorithm such as 0,1,2,3,4,5,6.

If(input value)<4 then(output value)＝2：GoTo End

If(input value)>3 AND(input value)<6 then(output value)＝4 Else

(output value)＝6

End

Quantizer---the instrument or the algorithm of a kind of formation quantization (discrete form) information.

The result who differs from or subtract each other of residual quantity---two numerical value.

Such as, A=(B-C) expression A is the residual quantity of B and C.If B is greater than C, the residual quantity value is a positive number so.If C is greater than B, the residual quantity value is for negative.

Correlation---passing threshold or side-play amount are to the comparison of two numerical value.If two numerical value are by increasing or reducing certain threshold value and consistent, they can be called as relevant.

Coefficient correlation---be used for the threshold value or the deviate that compare in correlation.

Redundancy value---certain data in one group of information that is expressed as special value are called as redundancy, if it constantly repeats in these group data.

Such as, information sets={ 1,2,3,0,0,0,0,0,0,0,2}: numerical value 0 is a high redundancy value, and 2 are one slight redundancy is arranged.Can the be pressed row principle compression of this redundant information:

Numerical value Binary code The long code of variable bit

0 00 0

1 01 011

2 10 01

3 11 111

Need use the 2X11=22 position in order to use binary code recorded information group.

Variable code will be (1X7)+(2X2)+(2X3)=17. because the existence of redundant elements in information sets is collapsed into possibility.

YUV-Y is brightness, and UV is a color component.YUV also can represent by Y, Cr, three variablees of Cb.

YUV color coordinates---(being also referred to as Y, Cr, Cb) VICOMP uses the YUV color coordinates, so that utilize color component (U﹠amp by with respect to the Y sample U and V sample being carried out double sampling; V) intrinsic low frequency.Human eye also has different to the susceptibility of Y (brightness) and UV (color) component.

Frequency domain---most of information comprise video information, are at Shi Chengzhong, that is to say that the amplitude of information changed according to the time. frequency is the inverse of time: F=1/T, T=1/F.Information can be transformed into frequency domain from time domain by Fourier transform (FFT).Fast fourier transform is used in video coding, is exactly discrete cosine transform (DCT) such as the MPEG technology.In time domain, given one group of information (within a certain period of time) can obtain the amplitude of some.In the frequency city, given one group of information (in specific frequency domain bandwidth) can obtain the frequency of some.Because some vision signals contain a large amount of low frequency components more, a large amount of redundancies is arranged in frequency domain so can know information.

Fig. 1 realizes block diagram for the typical case of VICOMP encoder.The lower part of Fig. 1 shows common logic and memory unit (General Logic), they be in video coding and decoding usually and the parts that adopted as usual.To these common and adopted as usual parts, no longer be described such as memory, buffer (memory of another kind of form), counter, register and interface.The structure of these General Logic parts and function are known to the those of ordinary skill in logical design field.

Slightly different ordinary parts of this General Logic are " position wrapper (BitPacker).A position wrapper is the device that is used for will vary bit length coded combination to a bit stream that adjoins. computer data comprises fixed bit length composition, byte (8), word (16) and double word (32) usually.Because the output that the variable bit coding produces all lengths, so these codings must be by " packing " in the stream that adjoins.Carrying out the device " position wrapper " of this function does not describe at this as part of the present invention.

The going up of this block diagram (mainly) partly (arthmetic statement) shows video stream by the various processing of forming this video coding system. and these parts are parts of the present invention and describe in detail in this article. and these detailed descriptions comprise literal, logic flow diagram and logical statement.The algorithm of VICOMP can utilize suitable computer to be realized by software, this computer have the interface of input side and digitized video and at outlet side to the appropriate display device.The algorithm of this VICOMP also can specific hardware logic blocks realize, for example will be used in application-specific integrated circuit (ASIC) (ASIC) semiconductor like that.

The purpose of Y and UV filter is to reduce noise and improves signal to noise ratio.Any improvement to signal to noise ratio all will cause better coding result (more compression), because the noise that is included in the video is asynchronous (at random) with respect to video, the noise in being included in video can not be effectively incompressible.

Below in conjunction with accompanying drawing the specific embodiment of the present invention is described.

Non-linear differential quantizer

Neighbor Y (brightness) differential value of each pixel is one group according to 3 pixels and quantizes.Simultaneously, they are a group coding with 3 pixels also.Therefore be mould at the one-row pixels sum with 3.That is to say that sum of all pixels is wanted and can be divided exactly by 3 in the delegation.Such as, 9,12,639,720,1023 all can be used as effective quantity of one-row pixels.If the pixel quantity of digital video transducer output is not a mould with 3, such as, 640, encoder can be ignored pixel last in delegation so.It is irrational that every like this row reduces by 1 to 2 pixel.

For describing, can suppose that Y value (amplitude) is 7, the value of this expression from 0 to 127.For 7 bit value (+/-127 full-scale differential (full scaledifferential)), the output of quantizer can be+1 ,+5 ,+13 ,+26 ,+42 ,+59 ,+in 14 numerical value of 76 ,-1 ,-5 ,-13 ,-26 ,-42 ,-59 and-76 any one.Here there is not numerical value 0.Same principle can be applied in the figure place of other all lengths, and such as 5,6,8 etc., but the numerical value of use is according to the quantity of Y value figure place in this explanation, be calibrated suitably with different full-scale corresponding to one.For example, be used for 8 Y differential quantizer values (0 to 255 full scale value) will for+/-2,10,26,54,84,118 and 152.

The fractional value residual quantity

In ordinary video, can expect will have a large amount of indivedual residual quantity values less than+/-2 3 pixel groups, because the object occupation space usually will be considerably beyond 3 pixels in video image.Also available another method is explained this fact,, has a considerable amount of low frequency components in ordinary video that is.By merging the low amplitude value residual quantity, under situation about not damaging aspect any other of high fdrequency component or video quality, can obtain suitable first order compression.

If Y value originally less than+/-2, promptly-1,0 or+1, the output of quantizer can be-1 or+1.In ordinary video showed, naked eyes were to differentiate the very little pixel of those and adjacent difference.If come 3 neighbors of quantizer all to be+value in/-1, and if 3 and be-1, so these 3 numerical value be converted into-1 ,+1 ,-1, no matter and the numerical value before their why.Simultaneously, if come 3 neighbors of quantizer all to be+value in/-1, and they and be+1, then these 3 numerical value can be converted into+1 ,-1 ,+1, no matter the numerical value before their is why.So, 3 pixel groups and just be saved.Following table 1 has shown how quantizer is handled all less than+/-2 3 neighbors.

For 3 neighbors that come quantizer and be+3 and-3 situation, these values are not changed, but keep its original value.This point is very important in the series connection of these little residual quantity values.If change continuous+3 or-3, will cause the visual distortion of video medium and low frequency component.

Table I--initial value is less than+all 3 pixel groups of/-2

Initial value＜+/-2	Quantizer output	End value Type
			+1 +1 +1	+1 +1 +1	+1 +1 +1 A
+1 +1 0	+1 +1 +1	+1 +1 +1 A
			+1 +1 -1	+1 +1 -1	+1 -1 +1 C
+1 0 +1	+1 +1 +1	+1 +1 +1 A
			+1 0 0	+1 +1 +1	+1 +1 +1 A
+1 0 -1	+1 +1 -1	+1 -1 +1 C
			+1 -1 +1	+1 -1 +1	+1 -1 +1 C
+1 -1 0	+1 -1 +1	+1 -1 +1 C
			+1 -1 -1	+1 -1 -1	-1 +1 -1 B
0 +1 +1	+1 +1 +1	+1 +1 +1 A
			0 +1 0	+1 +1 +1	+1 +1 +1 A
0 +1 -1	+1 +1 -1	+1 -1 +1 C
			0 0 +1	+1 +1 +1	+1 +1 +1 A
0 0 0	+1 +1 +1	+1 +1 +1 A
			0 0 -1	+1 +1 -1	+1 -1 +1 C
0 -1 +1	+1 -1 +1	+1 -1 +1 C
			0 -1 0	+1 -1 +1	+1 -1 +1 C
0 -1 -1	+1 -1 -1	-1 +1 -1 B
			-1 +1 +1	-1 +1 +1	+1 -1 +1 C
-1 +1 0	-1 +1 +1	+1 -1 +1 C
			-1 +1 -1	-1 +1 -1	-1 +1 -1 B
-1 0 +1	-1 +1 +1	+1 -1 +1 C
			-1 0 0	-1 +1 +1	+1 -1 +1 C
-1 0 -1	-1 +1 -1	-1 +1 -1 B
			-1 -1 +1	-1 -1 +1	-1 +1 -1 B
-1 -1 0	-1 -1 +1	-1 +1 -1 B
			-1 -1 -1	-1 -1 -1	-1 -1 -1 D

As above shown in the table 1, less than+/-2 3 pixel groups, there are 27 types for all initial residual quantity.This need use 5 to represent that these 27 kinds of whiles are at least with an extra group of discerning them.So obtain 6/3 pixel=2/pixel.Through after the quantizer, only have 8 types.Need to use 3 and represent that 8 all types need an extra bits to organize identification simultaneously.So obtain 4/3 pixel=1.33/pixel.But in end value, have only 4 types, this needs 2 to represent dissimilarly to add an identification group, so obtain 3/3 pixel=1/pixel.Compare with 7 original/pixel, this expression for all numerical value less than+/-2 3 pixel groups compression ratios are 7:1.

Referring to table 1, it should be noted that, there is 8 classes " A " output, 6 classes " B " output, 12 classes " C " output, and have only 1 class " D " output. because these little+/-1 neighbor residual quantity are usually by due to the white noise in the video, so quantize and make up these little residual quantity in fact to show as a kind of noise filter, reduce the quantity of different types of information simultaneously, therefore improved compression ratio.

This+/-1,3 neighbor residual quantity can further reduce by using the variable bit length coding based on the probability or the frequency of occurrences.Following table 2 has provided an example:

The variable bit length code of Table II+/-13 pixel groups

Type Code The possibility that takes place The average bit of 3 pixels

A 01 8/27＝0.296 0.592

B 011 6/27＝0.222 0.666

C 0 12/27＝0.444 0.444

D 111 1/27＝0.037 0.111

Total average bit/3 pixels 1.813

Add 1 and carry out type identification, the mean value of position/pixel will be (1+1.813)/3=0.937 position/pixel.7/pixel more originally, this expression for all values less than+/-2 3 pixel groups compression ratios are 7.47:1.Overall compression rate in the bright significant frame of this property list because in any typical video frame, can find 3 a large amount of pixel groups.

Fig. 2 illustrates the logical flow chart that little difference 3 pixel groups residual quantity quantize and make up.

Differential quantizer

Following Table III, IV, V illustrate non-linear differential quantizer of the present invention.In this example, the maximum scale value of supposing input information is 7 bit representation numerical value 0 to 127.The residual quantity of the Y that differential quantizer is adjacent with each, U and V value input is converted into 12 discrete steps (+or-) 1,5,13,26,52 and 76., and it is checked simultaneously and proofreaies and correct up and down and overflows---that is to say that combination that it does not allow to quantize residual quantity generates one greater than+127 or less than-127 maximum scale scope.

Differential quantizer of the present invention is permissible level 0 not.If continuously input value is a constant, will produce 0 continuous residual quantity at the input of quantizer.Yet for such input value, quantizer output will replace between+1 or-1, produce 1 error on the pixel that replaces.This mean error of 1/2 can become the part of quantizer noise.

Differential quantizer of the present invention is also as band pass filter, and its 3 decibels of cut-off frequencies are 1/2 of pixel sampling frequencies.Because the bandwidth of signal is limited, so the rise time is limited, they can not be in a sampling period from 0 to the maximum scale value.Maximum quantizer residual quantity 76 is based on that theoretical maximum slewing rate (slewing rate) that bandwidth is restricted to 1/2 signal of sample frequency selects.Because sample frequency must be at least the twice of the highest frequency component of signal, so differential quantizer can not limit the result's of this algorithm slewing rate.

Following Table III, IV, V have shown some examples of the quantification residual quantity of adjacent Y pixel value:

The Y pixel	Input Y value	Residual quantity (Y of input-Xin)	Quantizer output	New Y value	Mistake=(the Y value of Y value-Xin)
						The Yn-1 initial value	10	-	-	10	-
Yn	16	+6	+5	15	-1
						Yn+1	29	+14	+13	28	-1
Yn+2	84	+56	+59	87	+3
						Yn+3	89	+2	+1	88	-1
Yn+4	90	+2	+1	89	-1
						Yn+5	90	+1	+1	90	0
Yn+6	90	0	+1	91	+1
						Yn+7	90	-1	-1	90	0

Table III explanation quantizer is for the response of the six steps variation of input from 10 to 90

The Y pixel	Input Y value	Residual quantity (Y of input-Xin)	Quantizer output	New Y value	Mistake=(the Y value of Y value-Xin)
						The Yn-1 initial value	10	--	--	10	--
Yn	90	+80	+76	86	-4
						Yn+1	90	+4	+5	91	+1
Yn+2	90	-1	-1	90	0
						Yn+3	90	0	+1	91	+1
Yn+4	90	-1	-1	90	0
						Yn+5	90	0	+1	91	+1
Yn+6	90	-1	-1	90	0
						Yn+7	90	0	+1	91	+1

Table IV explanation quantizer changes response for a step of input from 10 to 90

The Y pixel	Input Y value	Residual quantity (Y of input-Xin)	Quantizer output	New Y value	Mistake=(the Y value of Y value-Xin)
						The Yn-1 initial value	74	--	--	74	--
Yn	79	+5	+5	79	0
						Yn+1	85	+6	+5	84	-1
Yn+2	67	-17	·13	71	+4
						Yn+3	63	-8	-5	66	+3
Yn+4	73	+7	+5	71	-2
						Yn+5	88	+17	+13	84	-4
Yn+6	113	+29	+26	110	-3
						Yn+7	103	-7	-5	105	+2
Yn+8	109	+4	+5	110	+1
						Yn+9	95	-15	-13	97	+2
	Sum＝ 949			Sum＝951	Sum＝+2

Table V explanation quantizer is for the response of actual video sampling

Following Fig. 3 A-3C shows the logical flow chart of non-linear differential quantizer of the present invention.In the drawings, quantizer has dY and Pv input and QdY output, wherein, and dY=neighborhood pixels residual quantity, the last pixel value of Pv=, QdY=quantizer result.

Fig. 4 shows the schematic block diagram of the non-linear differential quantizer of the present invention of finishing above-mentioned difference quantification.Noting, is shown in the example though this block diagram is a Y component with pixel, and wherein principle and structure are equally applicable to the non-linear differential quantizer of UV component.As shown in the figure, have dY and Pv the input non-linear differential quantizer 20 comprise the first difference quantifying unit 10 and the second difference quantifying unit 14.Wherein, the first difference quantifying unit 10 comprises: quantize subelement 11, be used for each residual quantity of enter video information adjacent pixels input value is converted into a plurality of discrete steps, promptly, distribute a quantification output valve (corresponding to the step S11-S12 among Fig. 3 A) corresponding to each residual quantity, as mentioned above, these discrete steps are calibrated full-scale with corresponding to pixel input range value suitably; Last overflow check subelement 12 and following overflow check subelement 13 are used for checking and proofread and correct up and down based on pixel input range value full-scale and overflow.If being checked through, last overflow check subelement 12 overflows, then next smaller step size value is assigned as new quantification output valve and proceeds overflow check (corresponding to the step S13 among Fig. 3 A, S15-S16), otherwise the quantification output valve of exporting described distribution quantizes output (corresponding to the step S17 among Fig. 3 B) as first of the first difference quantifying unit 10.If detecting down, following overflow check subelement 13 overflows, then next smaller step size value is assigned as new quantification output valve and proceeds overflow check (corresponding to the step S14 among Fig. 3 A, S16), otherwise the quantification output valve of then exporting described distribution quantizes output (corresponding to the step S17 among Fig. 3 B) as first of the first difference quantifying unit 10.In addition, quantize to be input to the second difference quantifying unit 14 with described first, quantize output valve all less than under+/-2 the situation at first of continuous 3 neighbors, if these 3 first quantize the output valve sum is-1 o'clock, the second difference quantifying unit 14 will be exported corresponding to second of described 3 neighbors and quantize output-1, + 1 ,-1 as the final quantifications output of nonlinear quantizer 20, no matter and numerical value before their why (corresponding to step S52 among Fig. 2); And when these 3 first quantification output valve sums are+1, the second difference quantifying unit 14 will be exported corresponding to second of described 3 neighbors and quantize defeated+1,-1 ,+1 as the final quantifications output of nonlinear quantizer 20, no matter numerical value before their why (corresponding to step S53 among Fig. 2).Do not satisfy above-mentioned condition if first of neighbor quantizes output, then will corresponding first quantize output as nonlinear quantizer 20 final quantifications export (corresponding to step S53 among Fig. 2).

From the above description as can be known, redundant in order to increase with the improvement compression, VICOMP video encoder use nonlinear quantizer of the present invention, this quantizer quantizes the residual quantity of adjacent video information.Under the situation of luminance Y component, this quantizer uses the residual quantity of neighbor, and under the situation of color UV component, it uses adjacent 2 X, 3 block of pixels.Above-mentioned nonlinear quantization principle at the Y component is equally also applicable to the UV component.Do not repeat them here.This nonlinear quantizer of the present invention causes some loss in the amplitude size, but the mode that this loss produces is non-linear corresponding to human eye.This quantizer non-linear is designed to mate the non-linear of human eye responding system.Therefore, final loss in amplitude information is imperceptible for the observer.

On the contrary, the quantizer sampling frequency size of using in DCT (discrete cosine transform) coded system causes the loss of high-frequency information.The amount of this loss depends on the realization of quantizer.But, lacking some loss, compression is not good. the corresponding loss of high-frequency information has proved acutance and little details has been impacted.The loss of this high-frequency information can be awared for the observer, and can the deterioration video quality.

The present invention is described in detail in conjunction with embodiment; but those skilled in the art understand, and protection scope of the present invention is not limited to the details in the above-mentioned execution mode; under the prerequisite that does not deviate from principle of the present invention and spirit, also there are various distortion and change.Therefore, protection scope of the present invention is limited by following claim.

Claims (8)

1, a kind of non-linear differential quantizer that is used for video coding system comprises the first difference quantifying unit, and it comprises:

Quantize subelement, be used for each residual quantity of enter video information adjacent pixels input value is converted into a plurality of discrete steps, described discrete steps is calibrated full-scale with corresponding to pixel input range value suitably;

The overflow check subelement is used for checking and proofread and correct upper and lower overflowing based on pixel input range value full-scale;

Non-linear differential quantizer also comprises the second difference quantifying unit, it is configured to receive first of the described first difference quantifying unit and quantizes output, and when described first absolute value that quantizes output corresponding to more than one neighbor all is 1, then distribute corresponding to second of each pixel and quantize output as final quantizer output according to pre-defined rule, otherwise, quantize output as final quantizer output with described first.

2, non-linear differential quantizer as claimed in claim 1, wherein, the value of luminance Y component that described pixel input value is each pixel and color UV component, described quantification subelement is configured to distribute one to quantize output valve according to the residual quantity of adjacent pixels input value, and wherein, described overflow check subelement comprises overflow check subelement and following overflow check subelement, and it is configured to:

When the quantizer result corresponding to the residual quantity of neighbor input value is timing, the described overflow check subelement of going up is judged whether full-scale greater than described pixel input range value of last pixel input value sum in described quantification output valve and the neighbor, if greater than, the then described overflow check subelement of going up indicates described quantification subelement that next smaller step size value is assigned as new quantification output valve to proceed described overflow check, otherwise, then export described quantification output valve and quantize output as first of the first difference quantifying unit;

When corresponding to the quantizer result of the residual quantity of neighbor input value when negative, described overflow check subelement down judges whether last pixel input value sum is negative in described quantification output valve and the neighbor, if for negative, then descend the overflow check subelement to indicate described quantification subelement that next smaller step size value is assigned as new quantification output valve to proceed described overflow check, if not for negative, then export described quantification output valve and quantize output as first of the first difference quantifying unit.

3, non-linear differential quantizer as claimed in claim 2, wherein, if the continuous input Y component and the UV component value of neighbor are constants, then the input at described quantizer will produce 0 continuous residual quantity, described quantizer output alternately, produces 1 error on the pixel that replaces between+1 or-1.

4, non-linear differential quantizer as claimed in claim 1, wherein, described more than one neighbor is 3 neighbors, and wherein, the described second difference quantifying unit is configured to will quantize output corresponding to second of described 3 neighbors and be converted to-1 ,+1 when the first quantification output valve sum of these 3 neighbors is-1,-1, no matter and the numerical value before their why; And when first of these 3 neighbors quantize the output valve sum and be+1, will quantize output corresponding to second of described 3 neighbors to be converted to+1 ,-1 ,+1, no matter their numerical value before is why.

5, a kind of non-linear method to quantify difference that is used in the video coding system comprises step:

Each difference of adjacent pixels input value in the enter video information is converted into a plurality of discrete steps, and described discrete steps is calibrated full-scale with corresponding to pixel input range value suitably;

Check and proofread and correct up and down based on pixel input range value full-scale and overflow;

Also comprise step: receive through the described full-scale first quantification output of checking and proofreading and correct the step acquisition of overflowing up and down based on pixel input range value, and when described first absolute value that quantizes output corresponding to more than one neighbor all is 1, then distribute corresponding to second of each pixel and quantize output as final quantification output according to pre-defined rule, otherwise, quantize output as final quantification output with described first.

6, non-linear method to quantify difference as claimed in claim 5, wherein, the value of luminance Y component that described pixel input value is each pixel and color UV component, described residual quantity step of converting distributes one to quantize output valve according to the residual quantity of adjacent pixels input value, and wherein, described overflow check step comprises substep:

When the quantizer result corresponding to the residual quantity of neighbor input value is timing, judge whether full-scale greater than described pixel input range value of last pixel input value sum in described quantification output valve and the neighbor, if greater than, then next smaller step size value is assigned as new quantification output valve to proceed described overflow check, otherwise, then export described quantification output valve and quantize output as first;

When corresponding to the quantizer result of the residual quantity of neighbor input value when negative, judge whether last pixel input value sum is negative in described quantification output valve and the neighbor, if for negative, then next smaller step size value is assigned as new quantification output valve to proceed described overflow check, if, then export described quantification output valve and quantize output as first not for negative.

7, non-linear method to quantify difference as claimed in claim 6, wherein, if the continuous input Y component and the UV component value of neighbor are constants, then will produce 0 continuous residual quantity, described quantizer output alternately, produces 1 error on the pixel that replaces between+1 or-1.

8, non-linear method to quantify difference as claimed in claim 5, wherein, described more than one neighbor is 3 neighbors, and wherein, when the first quantification output valve sum of these 3 neighbors is-1, will quantizes output corresponding to second of described 3 neighbors and be converted to-1 ,+1,-1, no matter and the numerical value before their why; And when first of these 3 neighbors quantize the output valve sum and be+1, will quantize output corresponding to second of described 3 neighbors to be converted to+1 ,-1 ,+1, no matter their numerical value before is why.

Images