CN100576881C - A kind of self-adapting video image zooming engine based on polyphase filter technology - Google Patents

Abstract

The invention discloses a kind of self-adapting video image zooming engine based on polyphase filter technology, its inner basic structure comprises analysis of image content module, image data storage module, synchronous detecting generation module, horizontal scaling module and vertically scale module; The user can be provided with register and the control output signal parameter that register manipulation is controlled each module by the MCU interface, the interpolation filter coefficient is carried out online programming, thereby finish detection, storage and the interpolation calculation function of view data.

Description

A kind of self-adapting video image zooming engine based on polyphase filter technology

Technical field

The invention belongs to digital video image and handle and the Display Technique field, the design video image zooming engine is specifically related to a kind of self-adapting video image zooming engine based on multiphase filter.

Background technology

Video image zooming engine is a part indispensable in the video display processing system, it can carry out spatial alternation with the input picture of varied different-format resolution, amplify or narrow down to particular space resolution, to reach the purpose that on the display terminal of target size size, shows playback.Early stage image zoom equipment is based on linearity or cubic interpolation algorithm mostly, realizes that its interpolation effect is relatively poor though these interpolation methods are easy to hardware, and the image detail loss after interpolation is serious also usually with serious ring (ringing) noise.Along with develop rapidly and people's demand growing to the multimedia display product of digital video technology, the video image zooming technology is subjected to attention and application more and more widely.Meanwhile, the multimedia video field emerges various video format resolution day by day, as QVGA, WXGA, 1080P etc., and diversified display device, as smart mobile phone, MP4 player, large scale panel TV etc., all the video scaling engine is proposed how new requirement, how to have designed the research focus that video scaling engine that the numerous resolution convergent-divergents of a support can produce clear, as not have ringing noise image has simultaneously become each big video image display device manufacturer.

Polyphase filter technology is an important branch of Digital Signal Processing, this theory thinks that all band limit digital signals can change to new sample frequency through after interpolation, low-pass filtering and the extraction, selecting under the situation of suitable low pass filter, even can not lose original information ground recover the to make new advances signal of sample rate; Simultaneously, this theory also proposes a kind of hard-wired multiphase filter structure that is easy to, this technology is divided into some phase places with above-mentioned low pass filter and carries out filtering operation respectively, saved hardware spending, simplified Filter Design, the applicant is applied to the video image zooming field with this technology, has brought convenience for designing novel Graph Scaler.

Summary of the invention

The objective of the invention is to, a kind of high quality graphic zooming engine that can be used for various video display apparatus is provided.This Graph Scaler adopts a kind of adaptive interpolation method based on polyphase filter technology, at first picture material is analyzed, and carries out adaptive-interpolation according to the characteristic of each pixel then and calculates, and finally obtains clear muting image.Wherein the interpolation filter of core has adopted heterogeneous structure to organize design the most, and this filter calculates desired structure according to the change filter coefficient in output cycle, and this time-multiplexed structure is saved resource, is highly susceptible to hardware and realizes.Simultaneously, the invention allows for a kind of block-based parallel image data store organisation, after adopting this data organizational structure, invention is amplified level and is dwindled the unit and be integrated into unified arithmetic element, realized up and down scale transformation of image simultaneously with single filter, simplified the structure and saved the hardware resource expense.

In order to realize above-mentioned task, the present invention takes following technical solution scheme:

A kind of self-adapting video image zooming engine based on polyphase filter technology is characterized in that, this Graph Scaler internal structure comprises:

One Input Image Content analysis module, it comprises noise filter, edge detector and ring estimator, is used for inputted video image is carried out the noise removing preliminary treatment, edge analysis extracts rough edge and details, removes isolated noise point; Carry out the ringing noise prediction and calculation at last;

The filter of making an uproar is used for inputted video image is carried out the noise removing preliminary treatment, obtains clean stable image sequence and uses for the successive image check and analysis, and noise filter adopts non-linear Gaussian filter;

Edge detector is used for input picture is carried out edge analysis, extracts rough edge and details (image high-frequency information); Pass through a series of morphological operations then, the edge image that extracts is carried out further meticulous reprocessing, remove isolated noise point as edge corrosion, expansion;

The ring estimator is used for the ringing noise prediction and calculation;

One memory read/write control module, comprise write control module, the row deposit array and read control module, wherein:

Writing module is according to synchronously and in useful signal every row serially data based space structure being required writing line successively to deposit array deposits of inputted video image;

The control signal that read through model produces according to interpolation computation circuit is read required data concurrently successively from row is deposited, the row space of going forward side by side reorders;

Row is deposited array, and the input data are carried out spatial recomposition, realizes being input to and the conversion of line output and the conversion of data clock frequency from serial;

One level and vertically scale module based on heterogeneous structure, comprise coefficient storage unit and horizontal scaling filter respectively, coefficient storage unit and vertically scale filter, level wherein and vertical filter are mainly used in and input point is carried out interpolation calculation try to achieve output point; Level and vertical filter have adopted heterogeneous structure, and filter changes filter factor according to the output cycle, and it is comprehensive that the total length of level and vertical filter equals each phase lengths;

One input synchronous detecting generation module, comprise input synchronous detecting and output synchronous detecting, be mainly used in the sequential of control whole system with synchronously, this module is according to the spatial variations ratio of input and output image, with the generation of input signal cycle required memory control signal and line of input field synchronization useful signal;

Under the control of input sync signal, the control module of writing of image data storage module is written to row line by line with input image data and deposits in the array, and control cycle ground deposits the write port cycling at six row, and assurance effectively utilizes storage resources; Simultaneously, also will import data and send into the analysis of image content module, the noise filter of analysis of image content module carries out the noise reduction preliminary treatment to the image/video data of input earlier, removes the various noises that image is introduced in transmission course, obtains clean stable image sequence; By edge detector image is carried out edge and detail detection then, extract the image high-frequency information, carry out ringing noise by the ring estimator at last and estimate; Under the control of the output synchronizing signal that the synchronous detecting generation module generates, horizontal scaling module and vertically scale module are carried out filtering interpolation to the graph data of reading from the image memory module and are calculated, and the control signal that the employed coefficient of image memory module is produced by above-mentioned ringing noise estimator is adaptively in the coefficient storage unit selection.

Characteristics of the present invention are to have adopted adaptive interpolation method to carry out image interpolation to calculate, and this method is being greatly improved than conventional method aspect preservation image detail and the high-frequency information; Simultaneously, for realization level, vertical filter on hardware, adopted heterogeneous structure to realize cleverly, and proposed a kind of data storage scheme of effective saving line storage.

Compared with the prior art, technique effect of the present invention is embodied in:

1. the present invention adopts adaptive interpolation method to carry out image interpolation arithmetic, and this method makes full use of various contents of image and feature, can effectively protect high frequency information such as image border and details, makes processing back image more clear;

2. the present invention has adopted reconfigurable filter construction, but the coefficient online programming of filter can adjust according to different input videos source, obtain better zooming effect;

3. the hardware of median filter of the present invention realizes having adopted polyphase filter technology, cleverly filter is folded and the coefficient division, the periodic handoff factor table of control signal according to synchronous generation module generation, this designs simplification the complexity of realization of filter, effectively reduced the hardware resource expense;

4. the present invention has adopted the pre-processing module of content-based identification, can classify and discerns the content of image, produces content information and uses for post-processing module, makes that the interpolation algorithm calculating of back is more accurate;

5. the present invention proposes a kind of parallel data store organisation, this structure employing serial writes, the mode of parallel read-out, has solved the problem that the cross clock domain data interpolating calculates; Simultaneously, this structure sense data is a parallel mode, and follow-up interpolation calculation task can calculate in the same clock cycle, has saved the pipeline register, has improved system clock frequency;

6. the synchronous detecting of the present invention design and extraction module can have the synchronous and control signal of fixed phase difference according to input sync signal and the adaptive generation of zoom ratio and input signal, the control signal of using this module to produce, guarantee the synchronism of input/output signal and straightforward, avoided the memory read write conflict.

Description of drawings

Fig. 1 is the system configuration functional block diagram of video scaling circuit of the present invention;

Fig. 2 is image detection and pretreatment module algorithm workflow diagram;

Fig. 3 is a parallel data storage organization scheme schematic diagram;

Fig. 4 operates corresponding ordinal relation schematic diagram for storage read-write;

Fig. 5 is the schematic diagram of each picture element position and filter relation on the sef-adapting filter one dimension direction;

Fig. 6 is the heterogeneous structure figure of sef-adapting filter;

The present invention is described in further detail below in conjunction with drawings and Examples.

Embodiment

Referring to Fig. 1, the self-adapting video image zooming engine based on the multiphase filtering structure of the present invention, its inner basic structure comprises analysis of image content module, image data storage module, synchronous detecting generation module, horizontal scaling module and vertically scale module; The user can be provided with register and the control output signal parameter that register manipulation is controlled each module by the MCU interface, the interpolation filter coefficient is carried out online programming, thereby finish detection, storage and the interpolation calculation function of view data.

After vedio data is sent into system, at first enter the analysis of image content module, it comprises noise filter, edge detector and ring estimator, and the effect of this module is: at first analyze the high-frequency information that identifies in the two dimensional image, as edge, details etc.; Characteristic according to the human eye vision perception, human eye is only responsive to ring (Ringing) noise ratio that is positioned at smooth region in the image, therefore according to this principle, the analysis of image content module will estimate the smooth region that these may produce ringing noise according to detected high-frequency information prediction subsequently, and its mark come out, in order to control back filtering interpolation process.

The image data storage module has comprised that data write control module, memory read control module and line storage array three parts; This module has mainly been finished the buffer memory of input and output view data, tasks such as locus mapping transformation, the conversion of inputoutput data clock frequency and interpolated data reads, rearrangement.This module is the core of whole video Graph Scaler, and the sequential of module guarantees the stable of line storage read-write by the strict control of synchronous detecting generation module.

The synchronous detecting generation module is mainly used in the sequential of control whole system with synchronously, this module generates required output synchronizing signal and memory read write control signal automatically according to the synchronizing signal of input and the zoom ratio of user's setting, and the operation of control whole system is with synchronous.

The core that two modules of level and vertically scale are whole zooming engines, adaptive algorithm and heterogeneous structure filter thereof that two module application the present invention of level and vertically scale propose, input image data is carried out the space two-dimensional interpolation calculation, try to achieve corresponding each point pixel value in the output image, realized the arbitrary ratio convergent-divergent of video image.

As shown in Figure 2,, reduce the original The noise of image, established a noise filter in the analysis of image content module in order to improve the accuracy of detection.This noise filter uses the window of 5x5 size that image is carried out Gauss's noise reduction, this noise filter is selected Gauss's denoiser, the most spiced salt and class Gaussian noise in can the filtering picture signal obtains clean view data, improved the accuracy of follow-up detection greatly.

Edge detector has adopted the outstanding Canny edge detection operator of every index, and this algorithm can be regulated cut-off frequency according to the adaptive adjustment detection threshold of the property of the histogram of input picture, finally obtains complete detail of the high frequency.

Ring (Ringing) estimator has adopted recurrence morphology algorithm to carry out the computing of ringing noise location estimation.This algorithm carries out repeatedly morphology operations such as recurrence expansion and corrosion according to the marginal information binary map that edge detector generates, and estimates the topological structure at edge, carries out matching operation with edge labelling information then, obtains final noise region signature.

Conventional video convergent-divergent chip is separately realized convergent-divergent (amplification) two parts functional module on convergent-divergent under the level (dwindling) and the level mostly, main cause is because through behind the different scale transformations, the clock frequency difference of input, data clock can reduce when dwindling, amplification is that clock frequency promotes, and two process intermediate demands have metadata cache FIFO to carry out the clock conversion;

In order to change this traditional structure, save the logic and the storage resources of hardware circuit, the present invention proposes a kind of new parallel data storage organization, as shown in Figure 3, after adopting this storage organization, system can carve and read required interpolation point under unified clock frequency when needed, thereby saved extra buffer memory FIFO, simplified the structure of zooming engine.Specific implementation is as follows: the wall scroll line storage that with capacity is 2048 bits earlier folds by every unit 4 bit bit wides, forming the dual-port parallel-by-bit with 4 reading-writing port deposits, and then with 8 same line storages merging, composition has a write port, and the parallel-by-bit of 4 read ports is deposited array.After so organizing line storage, during the input data can be written to that each bar is capable and deposit by serial, parallelly at last from memory, read, be equal to the clock frequency that has improved data like this, realized the clock frequency conversion of data;

The read-write process that whole row is deposited is controlled by the synchronous detecting generation module.This module produces according to the sequential of input sync signal and phase information and writes row accordingly and deposit control signal, drives to write control module the video data of input is written to during row deposits in order; Simultaneously, the zoom ratio that this module is also set according to the user is calculated and is generated final capable field sync signal and read to go and deposits control signal, and the interpolating pixel point that driving the line storage read through model will need to use is read from row is deposited successively, sends into the Zoom module of back.

The read-write that row is deposited is seen shown in Figure 4 in proper order.In order to utilize row to deposit into the line data buffer memory and change the data clock frequency, the present invention adopts serial to write, the strategy of parallel read-out.Input data the writing during current line deposits of serial under the driving of writing control module by the address increment order, the writing line order is deposited middle sequential loop at 8 row; Under the driving of reading control module, desired data from depositing, is read row again then with the parallel data amount of each four lines while four pixels, pass through map addresses at last, it is the order of interpolation filter line of input and point that the spatial order of reading row and point is reset, and the computing module that is sent to the back carries out the computing of interpolation convergent-divergent.Write and read the address relationship mapping and follow the relation of following formula regulation:

n＝(m*M)/L

Wherein, m is the pixel coordinate of input point, and n is the pixel coordinate of output point, and M is following zoom ratio, and L is last zoom ratio.

After adopting above-mentioned parallel storage organization, the present invention becomes unified level, vertically scale filter with convergent-divergent on the level and following scaled set.Level of the present invention, vertically scale have used resample filter to carry out the interpolation arithmetic of pixel, and have adopted heterogeneous structure to carry out hardware circuit and realized.

Fig. 5 has represented in the whole interpolation process, filter coefficient and interpolating pixel spatial order relation.This relation is followed following mathematic(al) representation:

$y\left(m\right)=\underset{n=-\∞}{\overset{\∞}{\mathrm{\Σ}}}h(\mathrm{nL}+\mathrm{mM}\⊕L)x(\frac{\mathrm{mM}}{L}-n)$

Wherein x (n) is an input pixel data, and y (m) is an output pixel, and h (n) is the unit impulse response function of interpolation filter, zoom ratio under the M, and L is last zoom ratio.

The hardware configuration of level, vertical filter has adopted the multiphase filtering structure, as shown in Figure 6, the present invention is folded into a plurality of subfilters that length is R with 256 filter of length, and each interpolation calculation is used one group of filter factor to carry out linear weighted function and calculated, and obtains the pixel of corresponding phase; Filter calculates the index signal selecting filter phase place that generates according to the synchronous detecting generation module.

Claims (2)

1. the self-adapting video image zooming engine based on polyphase filter technology is characterized in that, this Graph Scaler internal structure comprises:

One Input Image Content analysis module, it comprises noise filter, edge detector and ring estimator, is used for inputted video image is carried out the noise removing preliminary treatment, edge analysis extracts rough edge and details, removes isolated noise point; Carry out the ringing noise prediction and calculation at last;

One image data storage module, comprise write control module, the row deposit array and read control module, wherein:

Write control module according to synchronously and in useful signal every row serially data based space structure being required writing line successively to deposit array deposits of inputted video image;

Read the control signal that control module produces according to interpolation computation circuit, concurrently required data are read from row is deposited successively, the row space of going forward side by side reorders;

Row is deposited array the input data is carried out spatial recomposition, realizes being input to and the conversion of line output and the conversion of data clock frequency from serial; It is earlier to be that the wall scroll line storage of 2048 bits folds by every unit 4 bit bit wides with capacity that described row is deposited array, forming the dual-port parallel-by-bit with 4 reading-writing port deposits, and then with 8 same line storages merging, composition has a write port, and the parallel-by-bit of 4 read ports is deposited array;

The one horizontal scaling module based on heterogeneous structure comprises coefficient storage unit and horizontal scaling filter, and wherein the horizontal scaling filter is mainly used in and input point is carried out interpolation calculation tries to achieve the horizontal output point; The horizontal scaling filter adopts the multiphase filtering structure, and the horizontal scaling filter changes filter factor according to the output cycle, and it is comprehensive that the total length of horizontal scaling filter equals each phase lengths;

The one vertically scale module based on heterogeneous structure comprises coefficient storage unit and vertically scale filter, and wherein the vertically scale filter is mainly used in and input point is carried out interpolation calculation tries to achieve the vertical output point; The vertically scale filter adopts the multiphase filtering structure, and the vertically scale filter changes filter factor according to the output cycle, and it is comprehensive that the total length of vertically scale filter equals each phase lengths;

Described multiphase filtering structure is to be that 256 filter is folded into a plurality of subfilters that length is R with length, and each interpolation calculation is used one group of filter factor to carry out linear weighted function and calculated, and obtains the pixel of corresponding phase; Filter calculates the index signal selecting filter phase place that generates according to the synchronous detecting generation module;

One input synchronous detecting generation module, comprise input synchronous detecting and output synchronous detecting, be mainly used in the sequential of control whole system with synchronously, this module is according to the spatial variations ratio of input and output image, with the generation of input signal cycle required memory control signal and line of input field synchronization useful signal;

Under the control of input sync signal, the control module of writing of image data storage module is written to row line by line with input image data and deposits in the array, and control cycle ground deposits the write port cycling at eight row, and assurance effectively utilizes storage resources; Simultaneously, also will import data and send into the Input Image Content analysis module, the noise filter of Input Image Content analysis module carries out the noise reduction preliminary treatment to the image/video data of input earlier, removes the various noises that image is introduced in transmission course, obtains clean stable image sequence; By edge detector image is carried out edge and detail detection then, extract the image high-frequency information, carry out ringing noise by the ring estimator at last and estimate; Under the control of the output synchronizing signal that the synchronous detecting generation module generates, horizontal scaling module and vertically scale module are carried out filtering interpolation to the graph data of reading from the image data storage module and are calculated, and the control signal that the coefficient of each Zoom module median filter is produced by above-mentioned ringing noise estimator is adaptively in the coefficient storage unit selection of correspondence; This is chosen process and follows following formula:

$y\left(m\right)=\underset{n=-\∞}{\overset{\∞}{\mathrm{\Σ}}}h(\mathrm{nL}+\mathrm{mM}\⊕L)x(\frac{\mathrm{mM}}{L}-n)$

Wherein x (n) is an input pixel data, and y (m) is an output pixel, and h (n) is the unit impulse response function of interpolation filter, zoom ratio under the M, and L is last zoom ratio.

2. the self-adapting video image zooming engine based on polyphase filter technology as claimed in claim 1 is characterized in that described noise filter adopts non-linear Gaussian filter.

Images