CN102256047B - Filtering-based 2*2 digital BINNING system - Google Patents

Abstract

The invention discloses a filtering-based 2*2 digital BINNING system, comprising a mode selecting module, a boundary processing module, a double cubic filtering module, a double linear filtering module and a column downsampling module, wherein the mode selecting module is used for determining a double linear filtering mode or a double cubic filtering mode, and outputting image data to the corresponding filtering module; the double cubic filtering module and the double linear filtering module respectively perform the double cubic filtering treatment and the double linear filtering treatment on the input image data and then output to the column downsampling module; and the column downsampling module downsamples the input image data in a column direction and then outputs the image data. The filtering-based 2*2 digital BINNING system solves the problems that the simple addition of merged pixels is performed by employing a linear array CCD (Charge Coupled Device) simulated BINNING technique to lead to the loss of a part of detailed information as well as the reductions of the contrast ratio and the MTF (Modulation Transfer Function) of remote sensing images.

Description

A kind of 2 * 2 digital BINNING systems based on filtering

Technical field

The invention discloses a kind of 2 * 2 digital BINNING systems based on filtering.

Background technology

Charge coupled device (Charge-Coupled Devices) is called for short CCD, is a kind of semiconductor imaging device based on photoelectricity transformation principle.The charge-trapping that it can produce incident photon and is exported collected electric charge by control circuit in its potential well in the mode that electric charge shifts, thereby formation is corresponding to the digital picture of optical imagery.Along with the development of semiconductor technology and fine technique is perfect, the ccd sensor manufacturing process technology improves constantly, and has occurred in recent years some new technologies on ccd sensor chip, and the BINNING technology is exactly one of them.The BINNING technology is by arranging consolidation circuit or merging software, a plurality of pixels are merged into the technology of a pixel, can effectively improve the frame speed of pixel response degree, signal to noise ratio and camera etc., but, this BINNING technology is combined pixel and carries out simple addition, and the lost part detailed information causes the decline of remote sensing images contrast and MTF.

What numeral BINNING algorithm adopted is the filtering interpolation method, this algorithm utilization two dimension continuous function carries out the filtering interpolation fitting, the multiple that merges according to pixel again requires to complete the coordinate inverse mapping from output pixel to the input pixel, in the good function of match, the inverse mapping point is resampled, thereby complete the calculated for pixel values of target pixel points.Filtering interpolation algorithm commonly used has bilinearity method and bicubic method, wherein, the effect of bicubic method is best, the high-frequency information that can keep preferably image, the image border is sharper keen, and details is more clear, has improved the problem of image blurring of bilinearity method, this algorithm is guaranteeing effectively to improve contrast and the MTF of image under the substantially constant prerequisite of signal to noise ratio.But in this hardware algorithm implementation procedure, amount of calculation is large, and storage and logical resource spending are many, and this hardware for algorithm realizes bringing very large difficulty.The bilinearity method is simple, and hardware realizes that easily, storage and logical resource are paid wages few, and still, this algorithm can lose the part high-frequency information, causes problem of image blurring.

Summary of the invention

Technology of the present invention is dealt with problems and is: overcome the deficiencies in the prior art, a kind of 2 * 2 digital BINNING systems based on filtering are provided.Employing the invention solves line array CCD device simulation BINNING technology and is combined pixel and carries out simple addition, and the problem of the decline of the lost part detailed information that causes and remote sensing images contrast and MTF.

Technical solution of the present invention is:

A kind of 2 * 2 digital BINNING systems based on filtering comprise: mode selection module, margin processing module, bicubic side's filtration module, bilinearity filtration module and row down sample module,

Mode selection module: receive the view data by the line array CCD generation of input, and when every even number line finishes, according to the model selection instruction deterministic model signal that receives, and mode signal and view data are outputed to margin processing module simultaneously;

Margin processing module: view data is carried out BORDER PROCESSING according to mode signal:

If mode signal is appointed as the bilinearity filter patterns, after according to bilinear interpolation, every row view data of input being processed, output to bicubic side's filtration module;

If mode signal is appointed as bicubic side's filter patterns, after according to one dimension bicubic side interpolation method, every row view data of input being processed, output to the bilinearity filtration module;

Bicubic side's filtration module comprises capable filtration module, row sampling module and row filtration module: when obtaining stable view data input, the row filtration module carries out the four lines view data of input outputing to capable sampling module after the filtered view data in bicubic side of line direction; Four lines view data after the row sampling module will be sampled outputs to the row filtration module; The row filtration module carries out outputing to the row down sample module after bicubic side's filtering of column direction to the four lines view data of input;

The bilinearity filtration module comprises capable filtration module, row sampling module and row filtration module: when obtaining stable view data input, the row filtration module carries out two row view data of input outputing to capable sampling module after the bilinearity filtering of line direction; Two row view data output row filtration modules after the row sampling module will be sampled; The row filtration module carries out outputing to the row down sample module after the bilinearity filtering of column direction to two row view data of input;

Row down sample module: receive the view data by bicubic side's filtration module or the input of bilinearity filtration module, view data is carried out export after the down-sampling of column direction.

The capable filtration module of described bicubic side filtration module carries out the filtering of line direction according to formula (1) to the image of inputting,

D[i，j]＝-x(1-x) ²d[i，j-1]+(1-2x ²+x ³)d[i，j]

(1)

+x(1+x-x ²)d[i，j+1]-x ²(1-x)d[i，j+2]

The row filtration module of described bicubic side filtration module carries out the filtering of column direction according to formula (2) to the view data of input,

D[i，j]＝-y(1-y) ²d[i-1，j]+(1-2y ²+y ³)d[i，j]

(2)

+y(1+y-y ²)d[i+1，j]-y ²(1-y)d[i+2，j]

Wherein, x, y be view data in the row direction with column direction on the coordinate spacing.

The capable filtration module of described bilinearity filtration module carries out the filtering of line direction according to formula (3) to the image of inputting,

D[i，j]＝(1-|x|)d[i，j]+|x|d[i，j+1] (3)

The row filtration module of described bilinearity filtration module carries out the filtering of column direction according to formula (4) to the view data of input,

D[i，j]＝(1-|y|)d[i，j]+|y|d[i+1，j] (4)

Wherein, x, y be view data in the row direction with column direction on the coordinate spacing.

The present invention compared with prior art has following advantage:

(1) existing BINNING technology realizes by analog circuit is set, the present invention is that the mode by digital filtering realizes that pixel merges, the present invention is taking full advantage of the useful information of all pixels, with the quality aspect that improves image, significant advantage is arranged.Existing BINNING technology carries out simply adding up saturated overflowing to pixel, can lose a large amount of detailed information, cause remote sensing images to the decline of comparison and MTF, bicubic method filtration module can keep the high-frequency information of image preferably, the image border is sharper keen, details is more clear, and this technology is guaranteeing effectively to improve contrast and the MTF of image under the substantially constant prerequisite of signal to noise ratio.The relatively existing BINNING technology of bilinearity method filtration module also can better keep useful information, and bicubic method filtration module reserved high-frequency information is slightly poor relatively, but, its hardware realization is easy, and storage and logical resource spending are few, is very convenient effective digital BINNING technical scheme.

(2) existing BINNING technology is merged into a pixel with a plurality of pixels in the cumulative mode of electric charge, can effectively improve the pixel response degree, the frame speed of signal to noise ratio and camera, the present invention merges into a pixel by digital filtering and sampling with a plurality of pixels, equally can be in the situation that guarantee that signal to noise ratio improves the frame speed of camera, to satisfy the requirement of subsequent data transmission, the invention has the advantages that by digital circuit BINNING technology more flexible, be easy to realize and high reliability, can realize by bicubic side's filtering algorithm and bilinearity filtering algorithm the BINNING system of N * N.

Description of drawings

Fig. 1 is structural representation of the present invention;

Fig. 2 is the mode selection module structural representation;

Fig. 3 is bicubic side's filtering modular structure schematic diagram;

Fig. 4 is bilinearity filtering modular structure schematic diagram;

Fig. 5 is the structural diagrams intention of down sample module;

Fig. 6 a～6d is result and corresponding Fourier transform result after each pattern of the digital BINNING of cameraman gray-scale map process;

Fig. 7 is that each digital BIN NING pattern Fourier transform result is listed as to the summation curve.

Embodiment

Below just by reference to the accompanying drawings the present invention is described further.

The principle of the invention

The image filtering interpolation is defined as the convolution of the continuous impulse response h (x, y) of discrete picture sampled point and 2-d reconstruction filter.

$s(x,y)=\underset{k}{\mathrm{\Σ}}\underset{l}{\mathrm{\Σ}}s(k,l)h(x-k,y-l)$

Wherein, s (x, y) is pixel to be inserted, and s (k, l) is the discrete sampling point.In order to reduce the algorithm complex of two-dimentional kernel interpolation, h (x, y) is resolved into x axle, the axial impulse response product of y.

h(x，y)＝h(x)h(y)

1. based on the digital BINNING algorithm of bicubic side's filtering

Bicubic side interpolation method x axle, the axial kernel function of y can be expressed as

$h\left(x\right)=\left\{\begin{array}{cc}(A+2){\left|x\right|}^3-(A+3){\left|x\right|}^2+1,& 0\&le;\left|x\right|<1\\ A{\left|x\right|}^3-5A{\left|x\right|}^2+8A\left|x\right|-4A,& 1\&le;\left|x\right|<2\\ 0& 2\&le;\left|x\right|\end{array}\right.$

$h\left(y\right)=\left\{\begin{array}{cc}(A+2){\left|y\right|}^3-(A+3){\left|y\right|}^2+1,& 0\&le;\left|y\right|<1\\ A{\left|y\right|}^3-5A{\left|y\right|}^2+8A\left|y\right|-4A,& 1\&le;\left|y\right|<2\\ 0& 2\&le;\left|y\right|\end{array}\right.$

Wherein, x, y are the coordinate spacing (equaling 1/2) of pixel, and A is adjustable parameter.

Can be obtained respectively on x axle and y direction of principal axis by above formula, under 2 * 2BINNING pattern, one dimension bicubic side interpolation algorithm formula is as follows:

D[i，j]＝-x(1-x) ²d[i，j-1]+(1-2x ²+x ³)d[i，j]

+x(1+x-x ²)d[i，j+11]-x ²(1-x)d[i，j+2]

D[i，j]＝-y(1-y) ²d[i-1，j]+(1-2y ²+y ³)d[i，j]

+y(1+y-y ²)d[i+1，j]-y ²(1-y)d[i+2，j]

2. based on the digital BINNING algorithm of bilinearity filtering

Bilinear interpolation x axle, the axial kernel function of y can be expressed as

$h\left(x\right)=\left\{\begin{array}{cc}1-\left|x\right|,& 0\&le;x<1\\ 0& \mathrm{otherwise}\end{array}\right.,$

$h\left(y\right)=\left\{\begin{array}{cc}1-\left|y\right|,& 0\&le;y<1\\ 0& \mathrm{otherwise}\end{array}\right.$

Wherein, x, y are the coordinate spacing of pixel.

Can be obtained by above formula, under 2 * 2BINNING pattern, one dimension bilinear interpolation algorithmic formula is as follows:

D[i，j]＝(1-|x|)d[i，j]+|x|d[i，j+1]

D[i，j]＝(1-|y|)d[i，j]+|y|d[i+1，j]

Wherein, x=y=1/2, D[i, j] for merging rear view data, d[i, j] be input image data.

Design

Be illustrated in figure 1 as structural representation of the present invention.Comprise: mode selection module, margin processing module, bicubic side's filtration module, bilinearity filtration module and row down sample module also are included in the memory cell that is used for carrying out buffer memory in the processing procedure of view data simultaneously.Below just describe respectively for concrete functional module.

1, mode selection module

The structure of mode selection module as shown in Figure 2, mode selection module receives in when work the view data that line array CCD produces.And receiving capable useful signal, after the view data of expression current line is effective, the model selection instruction that receives is judged the deterministic model signal when every even number line finishes.Mode signal and view data output to the margin processing module that is attached thereto simultaneously.

Judgment model instruction former when even number line finishes because: the model selection instruction may be upgraded at any time, but for guaranteeing the correctness of data, to the filtering of view data generally in the middle of the processing of data line and row be unalterable in the middle of processing, therefore, the model selection order that utilizes register-stored to receive, then whether the read mode selection instruction changes to detect preference pattern after every even number line finishes.

2, margin processing module

Margin processing module adopts pipeline system to carry out BORDER PROCESSING to view data.Undertaken by different mode according to mode signal:

When tupe is the bilinearity pattern, adopt bilinear interpolation that every row view data of input is processed.For with delegation's view data, when first view data input, directly export first view data; When second view data input, export the mean value of first view data and second view data; When the 3rd view data input, export the mean value of second view data and the 3rd view data; When the 4th view data input, export the mean value of the 3rd view data and the 4th view data.

When tupe is bicubic side's filter patterns, adopt one dimension bicubic side interpolation method that every row view data of input is processed.For with delegation's view data, when first view data input, four filter coefficients be multiply by first view data simultaneously, then output; When second view data input, first view data be multiply by coefficient-0.125 add that second view data multiply by coefficient 0.625 and add first view data and multiply by coefficient 0.625 and add that second view data multiply by coefficient-0.125, then output; When the 3rd view data input, first view data be multiply by coefficient-0.125 add that second view data multiply by coefficient 0.625 and add the 3rd view data and multiply by coefficient 0.625 and add that the 3rd view data multiply by coefficient-0.125, then output; When the 4th view data input, first view data be multiply by coefficient-0.125 add that second view data multiply by coefficient 0.625 and add the 3rd view data and multiply by coefficient 0.625 and add that the 4th view data multiply by coefficient-0.125, then output.

3, filtration module

Described filtration module comprises above-mentioned bicubic side's filtration module, bilinearity filtration module, and structure is respectively as shown in Fig. 3,4.Two filtration modules correspond respectively to bicubic side's filter patterns and bilinearity filter patterns.

Bicubic side's filtration module

Before the stable view data of acquisition was inputted, when delegation's view data input was only arranged, the row filtration module carried out the filtering processing after delegation's view data of inputting is copied four times after the formation four lines; When two row view data inputs was only arranged, the two row view data that the row filtration module will be inputted were carried out the filtering processing after copying respectively and once consisting of four lines; When 3 row view data whens input were arranged, after the row filtration module copies the third line view data of input once, consist of with three row view data of input and carry out filtering after four lines and process.(filtering of above-mentioned situation after having consisted of the four lines view data is processed with reference to following stabilized image data filtering processing method.)

When obtaining the input of four lines view data, reach stable view data input, the row filtration module carries out the four lines view data of continuous input outputing to capable sampling module after the filtered view data in bicubic side of line direction; Continuous four lines view data after the row sampling module will be sampled outputs to the row filtration module; The row filtration module carries out outputing to the row down sample module after bicubic side's filtering of column direction to the continuous four lines view data of input.

Under 2 * 2BINNING pattern, one dimension bicubic side interpolation algorithm formula is as follows:

D[i，j]＝-x(1-x) ²d[i，j-1]+(1-2x ²+x ³)d[i，j]

+x(1+x-x ²)d[i，j+1]-x ²(1-x)d[i，j+2]

D[i，j]＝-y(1-y) ²d[i-1，j]+(1-2y ²+y ³)d[i，j]

+y(1+y-y ²)d[i+1，j]-y ²(1-y)d[i+2，j]

Wherein, x=y=1/2, D[i, j] be view data after merging, d[i, j] be input image data, i, j distinguish the row, column number of data representing image.

The bilinearity filtration module

Before the stable view data of acquisition is inputted, when delegation's view data input is only arranged, after the row filtration module copies delegation's view data of inputting once, carry out filtering after formation two row.(filtering of above-mentioned situation after having consisted of two row view data is processed with reference to following stabilized image data filtering processing method.)

When obtaining two row view data inputs, reach stable view data input, the row filtration module carries out two row view data of continuous input outputing to capable sampling module after the bilinearity filtering of line direction; Continuous two row view data output row filtration modules after the row sampling module will be sampled; The row filtration module carries out outputing to the row down sample module after the bilinearity filtering of column direction to the continuous two row view data of input.

Under 2 * 2BINNING pattern, one dimension bilinear interpolation algorithmic formula is as follows:

D[i，j]＝(1-|x|)d[i，j]+|x|d[i，j+1]

D[i，j]＝(1-|y|)d[i，j]+|y|d[i+1，j]

Wherein, x=y=1/2, D[i, j] for merging rear view data, d[i, j] be input image data.

4, down sample module

As shown in Figure 5, down sample module is included in the capable sampling module of filtration module content and the row sampling module that is used for last output.

Capable sampling module under line direction mainly carries out data clock the down-sampling of 1/2 frequency division by a digit counter, it is input as view data, simultaneously in conjunction with row useful signal and data clock, be output as the data clock after view data, row useful signal and down-sampling.

Row sampling module under column direction mainly will be gone useful signal by a digit counter and be carried out 1/2 down-sampling, the data clock that it is input as after view data, row useful signal and down-sampling is output as capable useful signal after view data, down-sampling and the data clock after down-sampling.

Interpretation of result

Adopt the test result of bilinearity filtration module of the present invention and bicubic side's filtration module as follows:

The data that are used for test are the cameraman gray-scale map of 256 * 256 * 8Bit and the remote sensing images that actual remote sensing camera obtains 1536 * 1024 * 12Bit.For simply, objectively result images is estimated, adopt Fourier spectrum, Y-PSNR (PSNR), mean square error (MSE) and four indexs of MTF to analyze here.

Be the normal bilinearity filter patterns of cameraman gray-scale map BINNING result as Fig. 6 (a); Fig. 6 (b) is cameraman gray-scale map bicubic side filter patterns BINNING result; Fig. 6 (c) is the Fourier spectrum figure of bilinearity filter patterns BINNING result; Fig. 6 (d) is the Fourier spectrum figure of bicubic side filter patterns BINNING result; Each Fourier transform is removed DC component.The relative bilinearity filter patterns of Output rusults that obviously can find out bicubic side's filter patterns from result remains with more high-frequency information.For the more comparison of quantification, each pattern Fourier transform result is listed as to summation obtains curve as Fig. 7, wherein, the sum1 curve is that bilinearity filter patterns Fourier transform result is listed as to summed result, and sum2 is that bicubic side's filter patterns Fourier transform result is listed as to summed result.

Be Y-PSNR, mean square error and the MTF result of the cameraman gray-scale map of the remote sensing images of three 1536 * 1024 * 12Bit and 256 * 256 * 8Bit as table 1.As can be seen from the table, the mean square error of bicubic side's interpolation result and mtf value are greater than bilinearity filter patterns result of calculation, be that the detailed information that the method keeps wants many, can effectively improve contrast and the MTF of image, still, in bicubic side's filtering algorithm hardware implementation procedure, amount of calculation is large, storage and logical resource spending are many, therefore, from the complexity of algorithm and occupying of hardware resource, the bilinearity filtering algorithm is better than bicubic side's filtering algorithm.In a word, bicubic side's filtering algorithm and bilinearity filtering algorithm respectively have pluses and minuses, and the simple resource occupation of algorithm is few requiring, and when not high to required precision, select the bilinearity filtering algorithm; Requiring to keep the detailed information high-frequency information as far as possible, and do not minding that the complexity of algorithm and resource spending when many, select bicubic side's filtering algorithm.

Table 1 peak value is made an uproar ratio, mean square error and MTF result relatively

The unspecified part of the present invention belongs to general knowledge as well known to those skilled in the art.

Claims (3)

1. 2 * 2 digital BINNING systems based on filtering is characterized in that comprising: mode selection module, margin processing module, bicubic side's filtration module, bilinearity filtration module and row down sample module,

Mode selection module: receive the view data by the line array CCD generation of input, and when every even number line finishes, according to the model selection instruction deterministic model signal that receives, and mode signal and view data are outputed to margin processing module simultaneously;

Margin processing module: view data is carried out BORDER PROCESSING according to mode signal:

If mode signal is appointed as the bilinearity filter patterns, after according to bilinear interpolation, every row view data of input being processed, output to bicubic side's filtration module;

If mode signal is appointed as bicubic side's filter patterns, after according to one dimension bicubic side interpolation method, every row view data of input being processed, output to the bilinearity filtration module;

Bicubic side's filtration module comprises bicubic side's row filtration module, bicubic side's row sampling module and bicubic side's row filtration module: when obtaining stable view data input, bicubic side's row filtration module carries out the four lines view data of input outputing to bicubic side's row sampling module after the filtered view data in bicubic side of line direction; Four lines view data after bicubic side's row sampling module will be sampled outputs to bicubic side's row filtration module; Bicubic side's row filtration module carries out outputing to the row down sample module after bicubic side's filtering of column direction to the four lines view data of input;

The bilinearity filtration module comprises bilinearity row filtration module, bilinearity row sampling module and bilinearity row filtration module: when obtaining stable view data input, bilinearity row filtration module carries out two row view data of input outputing to bilinearity row sampling module after the bilinearity filtering of line direction; Two row view data output bilinearity row filtration modules after bilinearity row sampling module will be sampled; Bilinearity row filtration module carries out outputing to the row down sample module after the bilinearity filtering of column direction to two row view data of input;

Row down sample module: receive the view data by bicubic side's filtration module or the input of bilinearity filtration module, view data is carried out export after the down-sampling of column direction.

2. a kind of 2 * 2 digital BINNING systems based on filtering according to claim 1 is characterized in that: the capable filtration module of described bicubic side filtration module carries out the filtering of line direction according to formula (1) to the image of input,

D[i，j]＝-x(1-x) ²d[i，j-1]+(1-2x ²+x ³)d[i，j] (1)

+x(1+x-x ²)d[i，j+1]-x ²(1-x)d[i，j+2]

The row filtration module of described bicubic side filtration module carries out the filtering of column direction according to formula (2) to the view data of input,

D[i，j]＝-y(1-y) ²d[i-1，j]+(1-2y ²+y ³)d[i，j] (2)

+y(1+y-y ²)d[i+1，j]-y ²(1-y)d[i+2，j]

Wherein, x, y be view data in the row direction with column direction on the coordinate spacing; D[i, j] be view data after merging, d[i, j] be input image data, i, j distinguish the row, column number of data representing image.

3. a kind of 2 * 2 digital BINNING systems based on filtering according to claim 1 is characterized in that: the capable filtration module of described bilinearity filtration module carries out the filtering of line direction according to formula (3) to the image of input,

D[i，j]＝(1-|x|)d[i，j]+|x|d[i，j+1] (3)

The row filtration module of described bilinearity filtration module carries out the filtering of column direction according to formula (4) to the view data of input,

D[i，j]＝(1-|y|)d[i，j]+|y|d[i+1，j] (4)

Wherein, x, y be view data in the row direction with column direction on the coordinate spacing; D[i, j] be view data after merging, d[i, j] be input image data, i, j distinguish the row, column number of data representing image.

Images