CN1901602A - Quick double linear interpolating method in image amplification process - Google Patents

Abstract

This invention discloses a quick dual-linearity interpolation method used in image amplification, which first of all interpolates directly in the coordinate system on an original image, then maps from the original to the target one so as to reduce the traversal operation in the process of the image amplification, besides, border pixels can be processed in a unified program frame, the interpolation is carried out on the line and row directions separately and redundant interpolation computation is removed and integer addition and subtraction and a integer shift operation are applied in the total interpolation operation, so that the velocity of amplifying images by dual-linearity interpolation is increased.

Description

The quick bilinear interpolation method that is used for image processing and amplifying process

Technical field

The present invention relates to a kind of interpolation method that is used for image processing and amplifying process, especially relate to a kind of quick bilinear interpolation method that is used for image processing and amplifying process.

Background technology

In the existing various video display apparatus, the image processing and amplifying is essential operation commonly used.For adapting to special occasions and obtaining visual effect preferably, usually need a kind of effective method to change the size of existing image, and the image after guaranteeing to change have preferable quality.For the processing and amplifying of image, not only requiring has extraordinary quality, also needs fast as far as possible speed.The simplest method is that translation repeats interpolation, promptly neighborhood pixels interpolation, and this method realizes that speed is very fast, but mosaic effect can appear in the image after amplifying.Bilinear interpolation method the most frequently used in the practical operation is fairly simple, and arithmetic speed is also than comparatively fast.But in some have application than higher requirement to arithmetic speed, when on high-resolution medical science dedicated display, handling big CR image, because this class image data amount is generally much larger than normal image, image processing needs more time, especially when multiplication factor is big obvious time lag can appear, in the similar this application, the processing speed that image amplifies becomes a bottleneck.

That generally adopts at present amplifies the bilinear interpolation method of handling to image, in two steps image is carried out geometric operation: at first, image is carried out space coordinate transformation; Secondly, image is carried out grey scale interpolation.And coordinate transform and grey scale interpolation all can exert an influence to the processing time that image amplifies, and in addition, grey scale interpolation also can have influence on last enlarged image quality.

For example, original image is the A image, and size is m * n; Target image after the amplification is the B image, and size is that (M＞m, N＞n), the amplification of the bilinear interpolation of image is divided into two steps to M * N: coordinate transform and bilinear interpolation.Coordinate transform is exactly to set up the coordinate mapping relations between the target image B behind original image A and the convergent-divergent.Bilinear interpolation is exactly on the basis of coordinate transform, calculates the color value of object pixel by four known pixels points that close on the object pixel point coordinates.As shown in Figure 1, four known adjacent pixels points are formed 2 little neighborhood squares, and we are called a processing unit to 2 such neighborhood squares.Each color level that is positioned at the target pixel points of processing unit inside can be carried out bilinear interpolation by the color level on four summits and be obtained (formula (1), formula (2)).

$\left.\begin{array}{cc}\left(i\right)& f\left(M\right)=a\×f\left(P_1\right)+(1-a)\×f\left(P_2\right)\\ \left(\mathrm{ii}\right)& f\left(N\right)=a\×f\left(P_3\right)+(1-a)\×f\left(P_4\right)\\ \left(\mathrm{iii}\right)& f\left(X\right)=b\×f\left(M\right)+(1-b)\×f\left(N\right)\end{array}\right\}...\left(1\right)$

(1) formula simplified obtains:

$\left.\begin{array}{cc}\left(i\right)& f\left(M\right)=f\left(P_2\right)+[f\left(P_1\right)-f\left(P_2\right)]\×a\\ \left(\mathrm{ii}\right)& f\left(N\right)=f\left(P_4\right)+[f\left(P_3\right)-f\left(P_4\right)]\×a\\ \left(\mathrm{iii}\right)& f\left(X\right)=f\left(N\right)+[f\left(M\right)-f\left(N\right)]\×a\end{array}\right\}...\left(2\right)$

(1), f (P) denotation coordination is the color value of P in (2) formula.

In interpolation convergent-divergent process, need all processing units of traversal, the number of processing unit has determined the traversal time, and the traversal time mainly is by pointer movement, and the expense of accessing operation is formed.At first the coordinate mapping relations to be between original image and target image, set up in the interpolation process, just interpolation can be carried out then.Usually the mapping method that adopts is that the target image coordinate is mapped in the original image.Be about in the target image each pixel coordinate (x ', y ') all by coordinate transform hint obliquely at relevant position in the original image (x, y):

$\left.\begin{array}{c}x=x^{\′}\×n/N\\ y=y^{\′}\×m/M\end{array}\right\}---\left(3\right)$

Calculate the value of pixel in the target image then according to small neighbourhood corresponding in the original image.As shown in Figure 2, (x is that pixel coordinate X '=(x ', y ') in the target image is mapped to the coordinate position (generally being decimal) in the original image y) to X=, form a processing unit with four pixels that X closes on, calculate its pixel value by bilinear interpolation.Calculate a target pixel points and need 3 floating-point multiplications, 6 floating adds.But because each target pixel points all will be mapped to a processing unit, the sum of processing unit has M * N.Simultaneously for the edge pixel of target image, exist X to drop on situation outside the original image, need handle especially this situation, carry out continuation such as edge pixel to original image.The expense that branch's jump instruction also will be arranged in processing procedure like this.Above-mentioned these factors combine, and have caused that present to carry out time of image processing and amplifying with bilinear interpolation method relatively long, and just the speed of Chu Liing is slower.

Summary of the invention

Technical problem to be solved by this invention is to propose a kind of quick bilinear interpolation method that is used for image processing and amplifying process, not influencing or seldom influence under the prerequisite of original image convergent-divergent quality, can improve the speed of processing greatly.

The present invention solves the problems of the technologies described above the technical scheme that is adopted: a kind of quick bilinear interpolation method that is used for image processing and amplifying process, for size is that original image and the size of m * n is the target image of M * N, definition original image and target image all with top left corner pixel as the origin of coordinates, right is the x positive direction, downward direction is the y positive direction, the coordinate of original image coordinate system mid point is (x, y), the target image pixel coordinate be (x ', y '), adopt following steps to obtain the target image gray values of pixel points then:

1) prepare on the column direction (x coordinate direction) of original image and line direction (y coordinate direction) respectively with step-length stepx, stepy carries out interpolation, stepx=(n-1)/(N-1); Stepy=(m-1)/(M-1) with the line number that j represents original image, gets j=0;

2) at the enterprising row interpolation of the line direction of original image, obtain in the original image coordinate system coordinate for (i ' * stepx, the gray value A of point j) (i ' * stepx, j), i '=0,1,, N-1 is the column number of enlarged image, and the interpolation result that obtains is deposited among the array Cur_Row, be Cur_Row[i ']=A (i ' * stepx, j);

3) line number of judgement original image, if the 0th row, promptly j=0 continues next step, if j ≠ 0 forwards 5 to);

4) value among the array Cur_Row is composed the 0th row, even B (i ', 0)=Cur_Row[i ' to target image], i '=0,1 ..., N-1 forwards 6 then to);

5) value of Cur_Row array is composed array, that is: Pre_Row[i ' to Pre_Row]=Cur_Row[i '], i '=0,1 ..., N-1;

6) make that the current line number of original image is j, with Pre_Row[i '], Cur_Row[i '] as [0,1] pixel value of Qu Jian two end points, be f (0)=Pre_Row[i '], f (1)=Cur_Row[i '], carry out linear interpolation with integer shift operation and integer addition and subtraction, obtain target image B OK, the grey scale pixel value of k row at i '

$B(i^{\′},k)=\tilde{f}(k\×\mathrm{stepy}-j+1);$

7) judge the line number of next image, j=j+1 is if j≤n-1 forwards 2 to), otherwise the end process process.Carrying out linear interpolation with integer shift operation and integer addition and subtraction is to calculate linear interpolation according to following false code

x∈[0，1]

int?k＝x×8+0.5

Switch(k)

{

$\mathrm{case}0:\tilde{f}\left(x\right)=f\left(0\right);\mathrm{break};$

$\mathrm{case}1:\tilde{f}\left(x\right)=f\left(0\right)+((f\left(1\right)-f\left(0\right))>>3);\mathrm{break};$

$\mathrm{case}2:\tilde{f}\left(x\right)f\left(0\right)+((f\left(1\right)-f\left(0\right))>>2);\mathrm{break};$

$\mathrm{case}3:\tilde{f}\left(x\right)=((f\left(0\right)+f\left(1\right))>>1)-((f\left(1\right)-f\left(0\right))>>3);\mathrm{break};$

$\mathrm{case}4:\tilde{f}\left(x\right)=(f\left(0\right)+f\left(1\right))>>1;\mathrm{break};$

$\mathrm{case}5:\tilde{f}\left(x\right)=((f\left(0\right)+f\left(1\right))>>1)+((f\left(1\right)-f\left(0\right))>>3);\mathrm{break};$

$\mathrm{case}6:\tilde{f}\left(x\right)=f\left(1\right)+((f\left(0\right)-f\left(1\right))>>2);\mathrm{break};$

$\mathrm{case}7:\tilde{f}\left(x\right)=f\left(1\right)+((f\left(0\right)-f\left(1\right))>>3);\mathrm{break};$

$\mathrm{case}8:\tilde{f}\left(x\right)=f\left(1\right);\mathrm{break};$

}。

Mapping relations from the original image to the target image are: $\left\{\begin{array}{c}{x}^{\′}=x\×(N-1)/(n-1)\\ y^{\′}=y\×(M-1)/(m-1)\end{array}\right..$

Can be earlier at the enterprising row interpolation of line direction of adjacent two row of original image, then ordinate is carried out interpolation at the point between this two every trade number.

Compare with the existing Bilinear Method that is used for the image amplification, the present invention has four advantages:

(1) directly carries out interpolation operation in the coordinate system on original image earlier, from the original image to the target image, shine upon again, the Coordinate Calculation process in the original image of the Coordinate Conversion from target image not, make the traversing operation in the image processing and amplifying process reduce, reduced the image traversal time;

(2) from formula (1) as can be seen, therefore a pixel in the average every calculating target image of interpolation method originally just need carry out 3 sublinear interpolation, needs 3 * M * N interpolation altogether; The inventive method has been removed redundant computation, need only adopt (N-2) * m+ (M-2) * n interpolation arithmetic, and this advantage shows obvious more when multiplication factor is big more;

(3) from formula (3) as can be seen, the boundary pixel coordinate that target image may appear in mapping method originally is mapped to outside the scope of original image, x＞the N of i.e. (3) formula calculating, perhaps y＞M, so just need mapping to inject row and judge, if x＞N, perhaps y＞M, just need special processing (expanding), increased the expense of branch's jump instruction as pixel to original image; The mapping method that the present invention uses makes the boundary pixel coordinate of target image just in time be mapped on the border of original image, therefore can handle with unified structure, need not mapping relations are judged, simultaneously because boundary alignment makes linear interpolation arithmetic decreased number (four boundary pixels of target image only need interpolation arithmetic one time) make original unfavorable factor become present favorable factor.

What (four) original linear interpolation method adopted is floating-point operation, and linear interpolation arithmetic of the present invention carries out with integer-bit computing and integer addition and subtraction, has improved arithmetic speed.

Compare with the existing bilinear interpolation method that is used for image processing and amplifying process, adopting quick bilinear interpolation method of the present invention to carry out image amplifies, when the image multiplication factor was 2～10 times, arithmetic speed had improved 4～10 times, and speed-up ratio increases along with the increase of multiplication factor.

Description of drawings

Fig. 1 is the schematic diagram of bilinear interpolation method;

Fig. 2 is the schematic diagram of original mapping method;

Fig. 3 is the schematic diagram of the mapping method of the present invention's employing;

Fig. 4 removes the schematic diagram of redundant computation for the present invention;

Fig. 5 is the principle schematic of the quick interpolation method of bilinearity of the present invention;

Fig. 6 for the present invention and former method speed ratio than schematic diagram;

Fig. 7 is that the present invention is with respect to former method speed-up ratio schematic diagram;

Fig. 8 is the quick bilinear interpolation method flow chart that is used for the image amplification of the present invention.

Embodiment

Below in conjunction with accompanying drawing the present invention is described in further detail.

At this our branch mapping relations, remove redundant computation and fast three parts of interpolation describe.

Mapping relations:

The method that existing employing bilinear interpolation is carried out the image amplification is advanced row-coordinate conversion, carries out bilinear interpolation again.Promptly to each pixel coordinate of target image B (i ', j '), according to (3) formula be mapped to coordinate X=among the original image A (x, y), this moment x, y is generally non-integer.As shown in Figure 2, again according to (x, y) processing unit that four pixels that close on constitute, carry out bilinear interpolation according to (2) formula and calculate (x, y) gray value of locating, the pixel among target image B of every like this calculating all needs to seek corresponding processing unit (four pixels) among the original image A, in computational process, each processing unit all needs data are carried out accessing operation.Calculate all pixels among the target image B like this, need to seek M * N processing unit altogether.Simultaneously, according to (3) formula,, exist X to drop on situation outside the original image A for the edge pixel of target image B, need handle (carrying out continuation) to this situation especially, have the expense of branch's jump instruction during program realizes like this as edge pixel to original image A.

The present invention adopts mapping relations:

$\left\{\begin{array}{c}{x}^{\′}=x\×(N-1)/(n-1)\\ y^{\′}=y\×(M-1)/(m-1)\end{array}\right.---\left(4\right)$

This mapping is opposite with the mapping direction of existing method, is that (x y) is mapped to coordinate X ' among the target image B=(x ', y '), sees Fig. 3 for coordinate X=from original image A.Here be not that the point of the rounded coordinate among the original image A is shone upon, but to coordinate (x=i ' * stepx, y=j ' * stepy) shine upon, the set of these points is designated as Λ={ (x, y) | x=i ' * stepx, y=j ' * stepy, i '=0,1 ..., N-1; J '=0,1 ..., M-1}, the coordinate that Λ shines upon in target image B just in time be (x '=i ', y '=j '), i '=0,1 ..., N-1; J '=0,1 ..., M-1 is also promptly corresponding to all the pixel coordinates among the target image B.This mapping makes our directly each processing unit of traversal in original image A, with the coordinate points of this processing unit inside (x=i ' * stepx, the gray value A of y=j ' * stepy) (x=i ' * stepx, y=j ' * stepy) calculate the back directly compose give B (i ', j '), therefore need (m-1) * (n-1) individual processing unit altogether.Simultaneously because shine upon from original image A to target image B, the boundary pixel coordinate of original image A just in time is mapped on the border of target image B, (m-1) * (n-1) individual processing unit has exactly covered target image B like this, edge pixel can not occur and fall situation outside the processing unit.

Remove redundant computation:

The bilinear interpolation that adopts (2) formula to represent, the calculating of each object pixel B all needs six floating adds and 3 floating-point multiplications, is actually some here and calculates redundant.(i) in point (2) formula that lateral coordinates is identical in same processing unit, (ii) two interpolation calculation are identical, just calculating difference (iii).As being X to coordinate among Fig. 4 ₁With X ₂The calculating of color level can use Q ₁, Q ₂2 color level interpolation obtains.It is public that two neighbouring processing units have a limit, and it is identical therefore an interpolation arithmetic being arranged also, as X ₂And X ₂Q is all used in the calculating of 2 color values ₂Color level.Adopt (4) formula mapping mode to make us can in interpolation process, remove these redundant computation.

In order to remove redundant computation, define two arrays, Pre_Row and Cur_Row store the data that interpolation obtains on the original image A line direction temporarily, Cur_Row stores the data that current j row interpolation obtains, and Pre_Row storage front delegation is the data that the j-1 row interpolation obtains.On column direction, carry out interpolation then for ordinate all coordinate points between j-1 and j among the set Λ.After this process is finished, value in the Cur_Row array is composed the array to Pre_Row, with j+1 behavior current line, recomputate the Cur_Row array, utilize Pre_Row and Cur_Row on column direction, to carry out interpolation for ordinate all coordinate points between j and j+1 among the set Λ.Above this process constantly repeat all row in handling original image A, i.e. j=n-1.By the interpolation arithmetic on separate rows direction and the column direction, we can remove those redundant computation steps like this.Improve the speed of image processing and amplifying.In addition, first element of Cur_Row array and last element do not need to carry out interpolation arithmetic, directly can be obtained by the pixel value assignment among the original image A, so need interpolation (N-2) * m time on the line direction altogether; When interpolation calculation is gathered the gray value of coordinate points among the Λ on column direction, for ordinate is the coordinate points of 0 and (N-1) * stepy=n-1, directly the analog value of composing with the Cur_Row array gets final product, therefore need interpolation (M-2) * N time on the column direction altogether, the number of the interpolation arithmetic that needs altogether of image is (N-2) * m+ (M-2) * N; And if (M-1) % (m-1)=0; (N-1) % (n-1)=0, then all pixels of original image A are just in time hinted obliquely on the rounded coordinate point of B, therefore have m * n grey scale pixel value indirect assignment to get final product.The pixel number that needs interpolation arithmetic to calculate gray value among the B can further reduce to M * N-m * n, and promptly each pixel among the target image B is average only need just can obtain less than the once linear interpolation.

For shown in Figure 4, Pre_Row[k] memory point Q ₁Gray value, Cur_Row[k] storage Q ₂The gray value of point, X ₁And X ₂Be processing unit P ₁, P ₂, P ₃, P ₄Be arranged in the point of set Λ in the inside, and abscissa is k * stepx.Pass through Pre_RoW[k] and Cur_Row[k] interpolation calculation X ₁And X ₂The point gray value, then make Pre_Row[k]=Cur_Row[k], recomputate Cur_Row[k] value be Q ₃Gray value, pass through Pre_Row[k] and Cur_Row[k] interpolation calculation X ₃And X ₄The gray value of point.

Quick interpolation:

In image interpolation arithmetic, the floating-point operation and the multiplying of interpolating function are more consuming time, and the color value of image is a positive integer, and the floating number that interpolation calculation obtains finally also will be converted into integer.The shift operation of integer and add operation are fast more a lot of than floating-point operation, if can all be converted into integer shift operation and add operation to floating-point operation and multiplying, then can reduce operation time.

If but interpolating function is a single order derived function f (x), x ∈ [0,1] (in bilinear interpolation, f (x) is a linear function), the situation of consideration interpolation.If the functional value at two end points places of line segment is known, the y value of 1 x correspondence in the middle of desiring to ask, then y=f (x) because x is a floating number, calculates f (x) and need carry out floating-point operation, and floating-point operation is more consuming time with respect to integer arithmetic.As shown in Figure 5, [0,1] interval is divided into 2 ⁿFive equilibrium is established the cut-point nearest apart from x and is

K ∈ [0,2 ⁿ], for given little positive number ε, if cut-point

Enough little with the distance of x, by f (x), the continuity of x ∈ [0,1] then has $|f\left(x\right)-f\left(\frac{k}{2^n}\right)|<\mathrm{\&epsiv;}.$ If

Calculating can obtain by add operation and shift operation, avoid floating-point operation and multiplying, must save operation time.We round the result who obtains again and are designated as carrying out accurate Calculation with floating number earlier With

The result that approximate f (x) obtains is designated as

$\tilde{f}\left(x\right)=f\left(\frac{k}{2^n}\right).$

Because the color level of image is limited, the floating number that interpolation arithmetic will be obtained at last forces to be converted to integer, and this has just caused round-off error, if get rounding, promptly Then rounding the mean error that causes is 0.25, and rounding the worst error that causes is 0.5, is the image of L for gray scale, and a pixel will adopt log ₂ ^LThe position is represented, uses

The signal to noise ratio of approximate f (x) is:

${\mathrm{SNR}}_{\hat{f}}10{\log }_2^{\frac{L^2}{{\left(0.5\right)}^2}}=20{\log }_2^{\frac{L}{0.5}}=20({\log }_2^L+1)---\left(5\right)$

When $|f\left(x\right)-f\left(\frac{k}{2^n}\right)|\&le;\mathrm{\&epsiv;},$ Then

${\mathrm{SNR}}_{\tilde{f}}=10{\log }_2^{\frac{{\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="A20061005238300141.png"\; state="\{\{state\}\}">L</figure-callout>}}^2}{{\mathrm{\&epsiv;}}^2}}=20{\log }_2^{\frac{L}{\mathrm{\&epsiv;}}}=20({\log }_2^L+{\mathrm{<figure-callout\; id="2"\; label="log"\; filenames="A20061005238300141.png"\; state="\{\{state\}\}">log</figure-callout>}}_2^{\frac{1}{\mathrm{\&epsiv;}}})---\left(6\right)$

When ε=0.5, ${\mathrm{SNR}}_{\hat{f}}={\mathrm{SNR}}_{\tilde{f}}$

Under the situation of linear interpolation:

f(x)＝(1-x)·f(0)+x·f(1)＝(f(1)-f(0))·x+f(0) (7)

$\left|f(x-\frac{k}{2^n})\right|\&le;|f^{\&prime;}\left(\mathrm{\&xi;}\right)\&CenterDot;(x-\frac{k}{2^n})|\&le;|f\left(1\right)-f\left(0\right)|\&CenterDot;\frac{1}{2^{n+1}}---\left(8\right)$

So for given ε＞0, as long as get

$N={\log }_2^{(f\left(1\right)-f\left(0\right))}+{\mathrm{<figure-callout\; id="2"\; label="log"\; filenames="A20061005238300141.png"\; state="\{\{state\}\}">log</figure-callout>}}_2^{\frac{1}{\mathrm{\&epsiv;}}}-1---\left(9\right)$

When n 〉=N

Just have: $|f\left(x\right)-f\left(\frac{k}{2^n}\right)|<\mathrm{\&epsiv;}$ Set up.

Carrying out in the image zoom process with bilinear interpolation, f (0), the gray value of adjacent two pixels of f (1) expression, by the local continuity of image, these two gray values differ generally less, unless 0 impulsive noise is arranged in very tangible border or the image.Therefore in most cases, | f (1)-f (0) |＜16=2 ⁴If, getting ε=0.5, just can guarantee ${\mathrm{SNR}}_{\tilde{f}}={\mathrm{SNR}}_{\hat{f}},$ By (9) formula, obtain N=4.According to the weber law, | f (1)-f (0) | when big, human eye perceives also can increase to differentiated threshold, so ε is slightly bigger also can not discovered by human eye.Can in image bilinear interpolation convergent-divergent, get n=4, adopt $\tilde{f}\left(x\right)=f\left(\frac{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="A20061005238300141.png"\; state="\{\{state\}\}">k</figure-callout>}}{2^4}\right)$ Than adopting

The snr loss who causes is negligible, if Can calculate fast and effectively, just can increase substantially the operational performance of bilinear interpolation Zoom method. $\tilde{f}\left(x\right)=f\left(0\right)+[f\left(1\right)-f\left(0\right)]\&times;\frac{\mathrm{<figure-callout\; id="16"\; label="k"\; filenames="A20061005238300152.png"\; state="\{\{state\}\}">k</figure-callout>}}{16},$

Can be expressed as Addition or the combination of subtraction because $[f\left(1\right)-f\left(0\right)]\&times;z,z=\frac{1}{2},\frac{1}{4},\frac{1}{8},\frac{1}{16}$ Can realize by shift operation, so just can calculate by step 8)

The flow chart of the quick bilinear interpolation method that is used for image processing and amplifying process of the present invention as shown in Figure 8.

Below carry out the arithmetic speed that image amplifies with regard to the inventive method and describe.

Adopt 256 look lena original image gray level images of 256 * 256 sizes to amplify 1.5～21 times.The algorithm realization of under visual c++ 6.0 environment, programming, consider the influence of Compiler Optimization, the result of gained records under the release version, because what difference the interpolation image that former method and fast method obtain does not visually have substantially, test objective is in order to compare their time overhead.Fig. 6 and Fig. 7 have reflected former bilinear interpolation amplification method and comparison arithmetic speed of the present invention.Result from figure can see, the quick interpolation method that is used for image zoom that the present invention proposes is very effective for the efficient that improves existing processing method, along with the increase of multiplication factor, and the method operation efficiency that the present invention proposes even improved 20 times.In requiring than higher application to the image zoom processing method real-time, the present invention provides the approach of an extraordinary raising operation efficiency undoubtedly.

Claims (4)

1, a kind of quick bilinear interpolation method that is used for image processing and amplifying process, it is characterized in that for size being that original image and the size of m * n is the target image of M * N, definition original image and target image all with top left corner pixel as the origin of coordinates, right is the x positive direction, and downward direction is the y positive direction, and the coordinate of original image coordinate system mid point is (x, y), the target image pixel coordinate is (x ', y '), adopts following steps to obtain the target image gray values of pixel points then:

1) prepare on the column direction (x coordinate direction) of original image and line direction (y coordinate direction) respectively with step-length stepx, stepy carries out interpolation, stepx=(n-1)/(N-1); Stepy=(m-1)/(M-1) with the line number that j represents original image, gets j=0;

2) at the enterprising row interpolation of the line direction of original image, obtain in the original image coordinate system coordinate for (i ' * stepx, the gray value A of point j) (i ' * stepx, j), i '=0,1,, N-1 is the column number of enlarged image, and the interpolation result that obtains is deposited among the array Cur_Row, be Cur_Row[i ']=A (i ' * stepx, j);

3) line number of judgement original image, if the 0th row, promptly j=0 continues next step, if j ≠ 0 forwards 6 to);

4) value among the array Cur_Row is composed the 0th row, even B (i ', 0)=Cur_Row[i ' to target image], i '=0,1 ..., N-1;

5) value of Cur_Row array is composed array, that is: Pre_Row[i ' to Pre_Row]=Cur_Row[i '], i '=0,1 ..., N-1;

6) make that the current line number of original image is j, with Pre_Row[i '], Cur_Row[i '] as [0,1] pixel value of Qu Jian two end points, be f (0)=Pre_Row[i '], f (1)=Cur_Row[i '], carry out linear interpolation with integer shift operation and integer addition and subtraction, obtain target image B OK, the grey scale pixel value of k row at i '

$B(i^{\&prime;},k)=\tilde{f}(k\&times;\mathrm{stepy}-j+1);$

7) judge the line number of next image, j=j+1 is if j≤n-1 forwards 2 to), otherwise the end process process.

2, the quick bilinear interpolation method that is used for image processing and amplifying process as claimed in claim 1 is characterized in that carrying out linear interpolation with integer shift operation and integer addition and subtraction in step 6) is to calculate linear interpolation according to following false code

x∈[0，1]

intk＝x×8+0.5

Switch(k)

{

case?0： $\tilde{f}\left(x\right)=f\left(0\right);\mathrm{break};$

case?1： $\tilde{f}\left(x\right)=f\left(0\right)+((f\left(1\right)-f\left(0\right))>>3);\mathrm{break};$

case?2： $\tilde{f}\left(x\right)=f\left(0\right)+((f\left(1\right)-f\left(0\right))>>2);\mathrm{break};$

case?3： $\tilde{f}\left(x\right)=((f\left(0\right)+f\left(1\right))>>1)-((f\left(1\right)-f\left(0\right))>>3);\mathrm{break};$

case?4： $\tilde{f}\left(x\right)=(f\left(0\right)+f\left(1\right))>>1);\mathrm{break};$

case?5： $\tilde{f}\left(x\right)=((f\left(0\right)+f\left(1\right))>>1)-((f\left(1\right)-f\left(0\right))>>3);\mathrm{break};$

case?6： $\tilde{f}\left(x\right)=f\left(1\right)+((f\left(0\right)-f\left(1\right))>>2);\mathrm{break};$

case?7： $\tilde{f}\left(x\right)=f\left(1\right)+((f\left(0\right)-f\left(1\right))>>3);\mathrm{break};$

case?8： $\tilde{f}\left(x\right)=f\left(1\right);\mathrm{break};$

}。

3, the quick bilinear interpolation method that is used for image processing and amplifying process as claimed in claim 1 is characterized in that the mapping relations from the original image to the target image are: $\left\{\begin{array}{c}{x}^{\&prime;}=x\&times;(N-1)/(n-1)\\ y^{\&prime;}=y\&times;(M-1)/(m-1)\end{array}\right..$

4, the quick bilinear interpolation method that is used for image processing and amplifying process as claimed in claim 1 is characterized in that earlier the enterprising row interpolation of line direction at adjacent two row of original image, then ordinate is carried out interpolation at the point between this two every trade number.

Images