CN103167282B - A kind of online updating method of image compression dynamic regulation - Google Patents

Abstract

The present invention relates to a kind of online updating method of image compression dynamic regulation, in Wireless Image Transmission Systems, specifically realized the dynamic adjustment of compression of images by dynamic calculation amendment quantization parameter QP.The present invention carries out infra-frame prediction and inter prediction according to H.264 international standard to YUV image, it is achieved room and time dependency compresses；The QP value of the key reference frame of an image sequence is calculated according to image content information；According to the different information of key reference frame He other frame, calculate the energy threshold of other frame, determine the QP value of all frames in described image sequence；According to the QP value of frames all in described image sequence, described image sequence is carried out quantification treatment, completes image compression.The present invention can dynamically revise the video dynamic compression of quantization parameter QP according to image content information, completes the online updating of compression of images code check in Wireless Image Transmission Systems.

Description

A kind of online updating method of image compression dynamic regulation

Technical field

The present invention relates to a kind of online updating method of image compression dynamic regulation, specifically without line chart Image transmission system is realized the dynamic adjustment of compression of images by dynamic calculation amendment quantization parameter QP.

Background technology

Wireless Image Transmission Systems is the field that present applied research is more popular, wherein in coding and decoding video side In method, H.264 coding and decoding video is a kind of popular canonical algorithm, and this technology has higher compression ratio, The features such as applied range, have been widely used in the various fields such as video conference, real-time monitoring. It is that the image gathered is carried out the compression processes such as infra-frame prediction, inter prediction, entropy code, it is achieved image Compression storage.Use H.264 decoding technique that the compressed bit stream received is decoded in decoding end, recover former Beginning image.By H.264 algorithm, image can carry out higher compression to store, according to actual application need Seek the coding transmission of video.Quantifying during Video coding is the most important compression coding technology, It is by numerical value less for each image sampling point coding mapping one-tenth, determines reflecting of quantization by quantization parameter QP Penetrate relation, directly determine the code check of compression of images.Wireless transmitting system is all in quantization parameter QP definite value now In the case of be operated, it is impossible to enough in system operation, quantization parameter QP H.264 is carried out dynamically Adjust, it is achieved dynamic compression ratio change, it is impossible to realize effectively dynamically updating compression.

Summary of the invention

Shortcoming for existing wireless image transmission system, it is impossible to carry out dynamic corrections QP according to image content information, Realizing the deficiency that compression bit rate adjusts in real time, the technical problem to be solved in the present invention is to provide one can basis Image content information dynamically revises the video dynamic compression of quantization parameter QP, completes wireless image transmission system The online updating of compression of images code check in system.

The technical scheme is that a kind of image compression dynamic regulation of the present invention Line update method, comprises the following steps:

According to H.264 international standard, YUV image is carried out infra-frame prediction and inter prediction, it is achieved space and time Between dependency compression；

The QP value of the key reference frame of an image sequence is calculated according to image content information；

According to the different information of key reference frame He other frame, calculate the energy threshold of other frame, determine described The QP value of all frames in image sequence；

According to the QP value of frames all in described image sequence, described image sequence is carried out quantification treatment, completes Image compression.

Described key reference frame be in the I frame of this image sequence, i.e. this image sequence other images with reference to figure Picture.

Described according to image content information calculate an image sequence key reference frame QP value particularly as follows:

Calculate the image content information of key reference frame；

Set up empirical relation model；

It is calculated the QP value of key reference frame.

The image content information of described calculating key reference frame is particularly as follows: join by edge detection algorithm detection key Examine the gradient information of frame, set up macroblock edges direction histogram, obtain the most probable predictive mode of each macro block And reconstructed blocks, make to subtract by image, obtain the residual image close with actual coding.

Described macro block is 8 × 8 macro blocks.

Described empirical relation model is

R=I*QP*e^a

Wherein R represents code check, I representative image content information, i.e. residual image average:

$\underset{i=0}{\overset{M}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}{P}_{i,j}/M*N$

In formula, M is residual image abscissa, and N is the vertical coordinate of residual image, and QP represents quantization parameter, a=0.9 Representative model variable.

The described different information according to key reference frame He other frame, calculates the energy threshold of other frame, determines In described image sequence all frames QP value particularly as follows:

Calculate the energy threshold of other frame: Wherein α_iBy the filter used in motion prediction The energy of ripple device determines, N is the frame number of this image sequence；

Calculate the QP value of other frame: QP_t=Round (QP₀+ 51CurAvg), wherein QP0 is crucial ginseng Examining the QP value of frame, QPi is the QP value of other frame.

The invention have the advantages that

1. the dynamic compression of wireless image transmission system adjusts.This method is based on H.264 video coding and decoding technology, same Time combine image content information, the dynamic pressure of video image by dynamically amendment quantization parameter QP, can be completed Contracting, it is achieved during wireless image transmission, the most dynamically revise compression ratio, it is achieved compression ratio and figure picture element The Balancing selection of amount.

The most existing Wireless Image Transmission Systems only went to revise compression ratio system is out of service when, It is tested to the most again determine suitable systematic parameter.Online updating side by image compression dynamic regulation Method, it is possible to achieve dynamically adjust compression ratio in system operation, convenient under different channels and application demand Dynamically adjusting, run without halt system and reset, can meet various application to greatest extent needs Ask, thus there is extremely important practical significance and researching value.

Accompanying drawing explanation

The online updating method flow diagram of Fig. 1 image compression dynamic regulation of the present invention.

Detailed description of the invention

Below in conjunction with the accompanying drawings and embodiment the present invention is described in further detail.

The inventive method is as it is shown in figure 1, comprise the following steps:

1. by H.264 carrying out video compress

The YUV image that this embodiment is gathered by image capture device, uses the coding H.264 of standard Technology, carries out image in frame, inter prediction, is compressed image, it is achieved Video coding leading portion algorithm Compression.

2. calculate the QP value of key reference frame in one group of image sequence according to image content information；

(1) I two field picture content information calculates；

First with the gradient information of edge detection algorithm detection key reference frame I frame, 8 × 8 macroblock edges are set up Direction histogram, obtains each 8 × 8 pieces of most probable predictive modes and reconstructed blocks, makees to subtract by image, To the residual image close with actual coding；By the meansigma methods of residual image as the image content information of I frame,

$\underset{i=0}{\overset{M}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}{P}_{i,j}/M*N$

In formula, M is residual image abscissa, and N is the vertical coordinate of residual image.

(2) empirical relation model is set up

According to analysis of experimental data, in the case of quantization parameter is constant, image content information (residual image Average) and code check is inversely, both image content information was the most complicated, generated number of coded bits the biggest, code Rate is the highest.Otherwise, image is the simplest, and it is the fewest that coding generates bit number, and code check is the lowest.

In the case of image content information is fixing, quantization parameter and code check are the most inversely.If amount Changing coefficient the highest, code check also can be the highest.Otherwise, the lowest.

Show that discovery three exists obvious index linear relation by standard test sequences being done experiment, according to Technology of Data Fitting, the expression formula setting triadic relation is as follows:

R=I*QP*e^a

In formula, R represents code check, I representative image content information (residual image average), and QP represents and quantifies system Number, a representative model variable, bring test data into formula computation model parameter a=0.9 above.

Can be obtained by analysis of experimental data, the content information of quantization parameter QP and image be (residual image Average), and between code check, have empirical relation.

3. key reference frame and the different information of present frame, calculates energy threshold, determines QP value；

(1) calculating of the energy threshold of present frame

For only I frame and the image sequence of P frame, it was predicted that be unidirectional.The present invention because for image time Sequence requirement order display, has the requirement of real-time simultaneously, therefore only considers unidirectional situation, does not consider that B frame is double To prediction.Energy threshold is mainly determined by the energy of the wave filter used in motion prediction.Do at every two field picture After complete motion prediction, average energy change is calculated by following formula:

$\mathrm{CurAvg}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_i*i/N$

Wherein α_iFor corresponding filter energy, only consider unidirectional prediction due to the present invention, therefore take 0.5.

Definition CurAvg is the energy threshold of every two field picture, and it is current frame image and key reference two field picture Energy quality is divided by the gross energy of same image sequence.

(2) calculating of the energy threshold of present frame

Due in H.264 the span of regulation QP be 0～51 so, calculate present frame according to energy threshold QP energy threshold

QP_t=Round (QP₀+51CurAvg)

In formula, Qp0 is the Qp value of key reference frame, just can be calculated the QP of the present frame of correspondence by above formula Value, so that it is determined that all of QP value in image sequence.

4., according to the QP value of all frames in the image sequence calculated, carry out quantification treatment, complete compression coding；

According to quantitative criteria mapping equation

$\mathrm{FQ}=\mathrm{Round}\left(\frac{y}{Q_{\mathrm{step}}}\right)$

Wherein y is input sample point coding, and Qstep is for quantifying to compensate, and FQ is quantized value, and Round () is Bracket function.Quantization step Qstep has 52 values in h .264.Wherein QP quantization parameter is as amount Change the sequence number compensated.When QP goes the finest quantization of minima 0 interval scale, when QP takes maximum 51 Represent the most coarse quantization.QP often increases by 6, and step-length increases by 1 times.The operation expression of concrete quantizing process For

|Z_ij|=(| Wij | * MF+f) ＞ ＞ qbits

Claims (4)

1. the online updating method of an image compression dynamic regulation, it is characterised in that comprise the following steps:

According to H.264 international standard, YUV image is carried out infra-frame prediction and inter prediction, it is achieved space and time Between dependency compression；

The QP value of the key reference frame of an image sequence is calculated according to image content information；

According to the different information of key reference frame He other frame, calculate the energy threshold of other frame, determine described The QP value of all frames in image sequence；

According to the QP value of frames all in described image sequence, described image sequence is carried out quantification treatment, completes Image compression；

Wherein, image content information particularly as follows: with edge detection algorithm detection key reference frame gradient information, Set up macroblock edges direction histogram, obtain the most probable predictive mode of each macro block and reconstructed blocks, by figure As making to subtract, obtain the residual image close with actual coding, by the meansigma methods of residual image as key reference The image content information of frame；

Wherein it is determined that in described image sequence all frames QP value particularly as follows:

Calculate the energy threshold of other frame:Wherein α_iBy the filter used in motion prediction The energy of ripple device determines, N is the frame number of this image sequence, and i represents the serial number of image sequence；

Calculate the QP value of other frame: QP_t=Round (QP₀+ 51CurAvg), wherein QP₀For key reference The QP value of frame, QP_tQP value for other frame；

Described according to image content information calculate an image sequence key reference frame QP value particularly as follows:

Calculate the image content information of key reference frame；

Set up empirical relation model；

It is calculated the QP value of key reference frame；

The QP value of the described key reference frame calculating an image sequence according to image content information is particularly as follows: count Calculate the image content information of key reference frame, set up the QP value of image content information, code check and key reference frame Empirical relation model, R=I*QP*e^a, R represents code check, I representative image content information, QP representative amount Change coefficient, a representative model variable, believe according to the picture material of known code check and calculated key reference frame Breath is calculated the QP value of key reference frame.

The online updating method of a kind of image compression dynamic regulation the most according to claim 1, it is characterised in that Described key reference frame is the reference picture of other images in the I frame of this image sequence, i.e. this image sequence.

The online updating method of a kind of image compression dynamic regulation the most according to claim 1, it is characterised in that Described macro block is 8 × 8 macro blocks.

The online updating method of a kind of image compression dynamic regulation the most according to claim 1, it is characterised in that Described empirical relation model is

R=I*QP*e^a

Wherein R represents code check, I representative image content information, i.e. residual image average:

$\underset{i=0}{\overset{M}{\Σ}}\underset{j=0}{\overset{N}{\Σ}}{P}_{i,j}/M*N$

In formula, M is residual image abscissa, and N is the vertical coordinate of residual image, and QP represents quantization parameter, a=0.9 Representative model variable, P_i,jRepresent each pixel of residual image.