CN1286062C - Interpolation processing method for digital image - Google Patents

Abstract

The present invention provides an interpolation processing method for digital images. Firstly, a corresponding mapping point coordinate (x, y) of a picture element point of a target image in a source image is determined; then, gauss weighing Wi is determined according to a coordinate (m, n) of an adjacent picture element point of a mapping point; then, a value f(x, y) of the mapping point is calculated according to the gauss weighing Wi and a value Vi of the adjacent picture element; the interpolation is carried out through the gauss weighing, the process time of the interpolation is reduced, and simultaneously, the image quality is relatively enhanced.

Description

The interpolation process method of digital picture

Technical field

The present invention is a kind of disposal route of digital picture, and particularly a kind of Gauss's weight of utilizing is carried out the method for interpolation to digital picture.

Background technology

When size that changes image or converted image, original pixels will rearrange, and generates some new pixels then.When image amplified, the original pixels gap enlargement needed some new pixels to replenish the space, when image dwindled, pixel began to mix, and generates new pixel, in order to fill the space of generation, must be by calculating the color value of neighborhood pixels, and use this to be worth to remedy or the interpolation space.The method of being used when our said interpolation (Interpolation) that Here it is, interpolation method are image redistribution pixel, purpose is to determine the information of the space point between the known point in the image.

The interpolation method of traditional digital picture has a variety of, and wherein fastest is NearestNeighbor (nearest-neighbor), and this method only copies adjacent pixels simply, though this method speed is very fast, produces sawtooth through regular meeting; Bilinear (bilinear interpolation) method is by getting each pixel and its color average of four neighbor pixels up and down, and the shade of setting up appropriateness then between pixel seamlessly transits, and effect is better, but the time of cost is longer; Bicubic computing method (bicubic interpolation) is similar with Bilinear, and difference is that it is eight color of pixel mean values that calculate on each pixel and adjacent " rice " font direction, and this algorithm effect is best, but the most time-consuming.Above-mentioned several interpolation method all exists drawback, or poor effect, or consumed time is longer, therefore, after image transformation, how to utilize the short time to finish interpolation processing, to reduce the stand-by period of interpolation to image, be at present in Digital Image Processing, product be badly in need of and development company the direction emphasis that should make great efforts.

Summary of the invention

In view of this, the present invention addresses the above problem the interpolation method that proposes a kind of image, and it has improved the speed of handling again relatively when having improved quality of image processing.

The invention provides a kind of interpolation process method of digital picture, the pixel of at first determining target image in source images corresponding photosites coordinate (x, y); (m n), determines Gauss's weights W according to the coordinate of the neighbor pixel of this photosites then _iAt last according to this Gauss's weights W _iAnd the value V of this neighbor pixel _i, calculate this photosites value f (x, y).

According to the interpolation process method of digital picture provided by the invention, by Gauss's weight, four neighbor pixels carry out interpolation around only utilizing, and when reducing the interpolation processing time, have improved the effect of image again relatively.

Relevant detailed content of the present invention and technology, conjunction with figs. is described as follows:

Description of drawings

Fig. 1 is the overview flow chart of the interpolation process method of digital picture of the present invention;

Fig. 2 is the synoptic diagram of the image transformation of the interpolation process method of digital picture of the present invention;

Fig. 3 is the synoptic diagram that utilizes the neighbor point interpolation of the interpolation process method of digital picture of the present invention;

Fig. 4 is the weight table generative process process flow diagram of the interpolation process method of digital picture of the present invention.

Among the figure

The pixel that step 110 is determined target image in source images corresponding photosites coordinate (x, y).

(m n), determines Gauss's weight to step 120 according to the neighbor point coordinate of this photosites.

Step 130 is calculated the value of this photosites according to the value of this Gauss's weight and this neighbor pixel.

21: 21

22: 22

23: 23

24: 24

25: 25

26: 26

Step 410 is determined a zone between the neighbor pixel space, and determines a unique point in this zone.

(m n), calculates Gauss's weight to step 420 according to the neighbor point coordinate of this unique point.

This regional Gauss's weight of step 430 record.

Embodiment

Interpolation is exactly to determine in the image information of space point between the known point.We know that image is made up of pixel, when we amplify. dwindle or during image rotating, original pixels will rearrange, and generates some new pixels then.Therefore, just need reduce after the image geometry conversion destruction by interpolation to original image.

The present invention is described in detail below in conjunction with accompanying drawing.

Interpolation process method according to digital picture provided by the invention, see also Fig. 1, Fig. 1 illustrates method of the present invention, this figure is the interpolation process method overview flow chart of the digital picture that the present invention carried, the pixel of at first determining target image in source images corresponding photosites coordinate (x, y) (step 110) is according to the neighbor point coordinate (m of this photosites, n), determine Gauss's weights W _i(step 120) is according to this Gauss's weights W _iAnd the value V of this neighbor pixel _i, calculate value f (x, y) (step 130) of this photosites.

When we edit digital picture, through regular meeting image is done some geometric transformations, described geometric transformation is exactly the rotation to image. stretches. distortion. variations such as distortion.Through after the geometric transformation, pixel will rearrange, and after digital picture is carried out geometric transformation, certain reflection relation must be arranged between source images and the target image.Described reflection relation is description by the function of source images to target image, and the reflection relation has illustrated the conversion of being carried out from the source images to the target image.

See also Fig. 2 A, this figure is a source images, and four picture points are arranged among the figure, is respectively a little 21.: 22.: 23, and put 24, it is any four pixels in the image, after this image amplifies, obtain the target image among Fig. 2 B, at this moment, can set up the reflection relation of source images to target image.In target image, much newly-generated pixels as point 25.: 26 are arranged, need be to these newly-generated pixel assignment.After we have set up the reflection relation of source images and target image, just can find the point of newly-generated pixel correspondence in source images, calculate the value of giving newly-generated pixel by this point by reverse reflection.Described reverse reflection is from target image, can calculate photosites corresponding with it in the source images by each point in the target image.Divide from the direction of reflection, two kinds of basic mapping methods are arranged: forward reflection and reverse reflection.The forward reflection is from source figure, each point in the figure of source is sought photosites corresponding with it among the target figure, and inverse mapping is just in time opposite.In forward reflection, the same point in the target pattern of may videoing of a plurality of points among the figure of source causes double counting; And, if the reflection from source figure to target pattern is not full reflection, might make some point among the target figure can not get correspondence, form the cavity.Reverse reflection does not have these problems, therefore, has taked the mode of reverse reflection here.

In fact, the pairing function of reverse reflection, the just inverse function of forward reflection respective function.Suppose that Fig. 2 B is the image that the horizontal ordinate of the point among Fig. 2 A is all amplified correspondence after 2.5 times.For point 25, its reverse photosites is all dwindled its horizontal ordinate in Fig. 2 B 2.5 times of resulting points exactly so, and this point is corresponding photosites among source images Fig. 2 A.If the pairing point of reverse reflection is when just in time to be in the source images certain a bit, interpolation information is exactly the value of this point in the source images so.25 reverse reflection is pairing not necessarily can be found in source images but put, that is to say that the coordinate figure that reverse reflection obtains not necessarily just in time is the coordinate of pixel, if drop between the space of pixel, just need calculate the value of this point by the point around it.Here with this coordinate figure round numbers, the pairing pixel of this round values is exactly one of neighbor pixel, and as shown in Figure 3, (x y) is a little 23 photosites to some M, and the value of this photosites is interpolation information.With (x y) rounds the value of back gained for (i, j), therefore (i j) is one of neighbor pixel, and the coordinate of four consecutive point is respectively: M so on every side ₁(i, j), M ₂(i+1, j), M ₃(i, j+1), M ₄(i+1, j+1).Certainly the more point that also can stretch out, 16 points are as neighbor pixel around for example utilizing, and the number of neighbor pixel is many more, and the value of the photosites of being calculated is also just accurate more, but the time that consumes simultaneously also can be more.

After finding the pairing photosites of source images by reverse reflection, can utilize the adjacent point of this photosites to calculate the value of photosites.Here by the mode of the value weighting of source images being calculated the value of photosites.Utilize formula Gauss weights W _iAnd the value of neighbor pixel, calculate insertion point M (x, information y).Computing formula is:

$f(x,y)=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{W}_i*V_i$

Wherein, M _iBe photosites M (x, y) adjacent picture point, V _i(it just can obtain by imageing sensor from original image for x, the rgb value of each neighbor pixel y) for a M.Gauss's weight $W_i=e^{-r\sqrt{{(x-m)}^2+{(y-n)}^2}},$ And $\underset{i=1}{\overset{4}{\mathrm{\Σ}}}{W}_i=1,$ Wherein (m n) is the coordinate of each neighbor pixel, and r is a coefficient.

Repeatedly experiment shows that when the number of neighbor pixel was four, the time that is consumed was shorter relatively, and the effect of image is also pretty good, and four picture points are carried out interpolation calculation around therefore selecting here to utilize.When four pixels carry out interpolation around utilizing (the K value is 4), need be according to four neighbor pixel (M ₁, M ₂, M ₃, M ₄) coordinate Calculation go out corresponding four Gauss's weights, four Gauss's weights are respectively:

$W_1=e^{-r\sqrt{{(x-i)}^2+{(y-j)}^2}}$

$W_2=e^{-r\sqrt{{(x-i-1)}^2+{(y-j)}^2}}$

$W_3=e^{-r\sqrt{{(x-i-1)}^2+{(y-j-1)}^2}}$

$W_4=e^{-r\sqrt{{(x-i)}^2+{(y-j-1)}^2}}$

Make W ₁+ W ₂+ W ₃+ W ₄=1, so just can obtain the value of coefficient r, utilize the r value to return and calculate four weights.(x, value y) are the information that will insert just can to calculate f by above-mentioned four weights.

In above-mentioned interpolation process, what take is the method for photosites pixels all in the source images being carried out one by one interpolation calculation, in order to shorten the used time of interpolation, also can adopt the mode of looking into weight table to obtain weight, write down each area relative weight of image in the weight table, corresponding weights can be found in any zone in weight table, can calculate interpolation information by this weight.See also Fig. 4, this figure is a weight table generative process process flow diagram among the present invention.At first do what a grid, this grid is that the space between the pixel is subdivided into a plurality of blockages again, each blockage is a zone, and for example the grid that per four pixels are formed is further divided according to 100*100, just includes 10000 blockages between these four pixels so.For each piece blockage, any of selected its center is unique point (step 410), calculates weights W _iThe weight of this unique point is that (m, (step 420) n) calculated are noted (step 430) with coordinate and corresponding Gauss's weight of each piece again, are weight table according to its neighbor point coordinate equally.

When needing interpolation, can in weight table, find corresponding Gauss's weight by interpolated coordinates, and then according to this Gauss's weights W _iAnd the value V of this neighbor pixel _i, the value f of calculated characteristics point (x, y).The weights W here _iCalculating can be chosen in four such points and carry out, be (0,0) (0,1), (1,0), (1,1).Other situation this four points of can videoing calculate interpolation information according to this weighted value again, thereby have saved double counting many times.Handle by each piece blockage, in weight table, find its corresponding weights, and then calculate interpolation information.Thereby replace interpolation calculation one by one, reduced the used time of interpolation greatly each point.

Though the present invention discloses as above with aforesaid preferred embodiment; but it is not in order to limit the present invention; any people who has the knack of the camera image technology; without departing from the spirit and scope of the present invention; a little is changed and interpolation should to do some, and therefore scope of patent protection of the present invention is as the criterion with described in claims.

Claims (12)

1. the interpolation process method of a digital picture, the pixel that comprises the steps: to determine target image in source images corresponding photosites coordinate (x, y); (m n), passes through formula according to the coordinate of the neighbor pixel of this photosites

$W_i=e^{-r\sqrt{{(x-m)}^2+{(y-n)}^2}}$

Calculate Gauss's weights W _i, wherein r is a coefficient, its value is to pass through formula

$\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{W}_i=1$

Calculate; And

According to this Gauss's weights W _iAnd the value V of this neighbor pixel _i, calculate this photosites value f (x, y).

2. the interpolation process method of a digital picture comprises the steps:

The pixel of determining target image in source images corresponding photosites coordinate (x, y);

(m n), determines Gauss's weights W by looking into weight table according to the coordinate of the neighbor pixel of this photosites _iAnd

According to this Gauss's weights W _iAnd the value V of this neighbor pixel _i, calculate this photosites value f (x, y).

3. the interpolation process method of digital picture according to claim 1 and 2, wherein said definite photosites coordinate (x, step y) also comprises:

Determine the reflection relation from the source images to the target image; And

The coordinate of the reverse institute's corresponding point of videoing of the pixel of calculating target image.

4. the interpolation process method of digital picture according to claim 1 and 2, the number of wherein said neighbor pixel is 4.

5. the interpolation process method of digital picture according to claim 1 and 2, the number of wherein said neighbor pixel is 16.

6. the interpolation process method of digital picture according to claim 1 is wherein represented the number of described neighbor pixel with K, when the value of K is 4, if the coordinate of neighbor pixel be respectively (m, n), (m+1, n), (m, n+1), (m+1, in the time of n+1), W then _iValue be respectively:

$W_1=e^{-r\sqrt{{(x-m)}^2+{(y-n)}^2}}$

$W_2=e^{-r\sqrt{{(x-m-1)}^2+{(y-n)}^2}}$

$W_3=e^{-r\sqrt{{(x-m-1)}^2+{(y-n-1)}^2}}$

$W_4=e^{-r\sqrt{{(x-m)}^2+{(y-n-1)}^2}}$

Wherein r is a coefficient, and its value is to pass through formula

$\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{W}_i=1$

Calculate.

7. the interpolation process method of digital picture according to claim 6, (m n), is by (x y) rounds and obtains with the coordinate figure of photosites to the coordinate of wherein said neighbor pixel.

8. the interpolation process method of digital picture according to claim 2, the generation of wherein said weight table comprises the steps:

Between the space of neighbor pixel, determine a zone, and in this zone, determine a unique point;

(m n), calculates Gauss's weight according to the coordinate of this unique point neighbor pixel; And

Write down this regional Gauss's weight.

9. the interpolation process method of digital picture according to claim 8, the wherein said step of between the space of neighbor, determining a zone, be that each blockage is a zone for the area dividing that four neighbor pixels are formed is a plurality of blockages.

10. the interpolation process method of digital picture according to claim 8, wherein said unique point is the central point in this zone.

11. the interpolation process method of digital picture according to claim 1 and 2, wherein said value according to this Gauss's weight and this neighbor pixel, the step of calculating the value of photosites is to pass through formula:

$f(x,y)=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{W}_i*V_i$

Calculate, wherein K is the number of neighbor pixel.

12. the interpolation process method of digital picture according to claim 1 and 2, the value V of wherein said neighbor pixel _iRgb value for corresponding neighbor pixel.

Images