CN101001387B - Method of deciding image local information change minimum direction using multiple decision - Google Patents

Abstract

This invention relates to a method for utilizing majority decisions to judge the direction with the least image local information change including: getting multiple evaluation values of color information changed direction at an interested pixel place on at least two image directions and applying majority decisions to decide the direction with the least image local information change in the direction expressed by said multiple evaluation values, each value is taken from the difference value of the grads value of the color value at the interested image place or the flag value of the adjacent image of the interested image to express a trend of its color information change to increase the accuracy of judgement.

Description

Utilize majority rule to judge the method for image local information change minimum direction

Technical field

The present invention relates to a kind of method of judging the chromatic image local information change minimum direction, particularly about a kind of method of utilizing majority rule to judge the chromatic image local information change minimum direction.

Background technology

Utilize colorful optical filter array interpolation method (Color Filter Array Interpolation; CFAI) during the color information of reconstructing digital image, often need to judge image local information change minimum direction, the general method of using is to calculate pixels of interest Grad in different directions in the image, the size that compares Grad on the different directions again, the less direction of Grad is this image local information change minimum direction, and then carries out the required color-values of interpolation with the color-values of the neighborhood pixels on this direction.5 * 5 pel arrays 100 with Fig. 1 are example, and the R among the figure, G and B represent the color-values of redness, green and blue three kinds of colors respectively, and its subscript is represented the coordinate of this location of pixels.At the green tint value G that rebuilds pixel (2,2) ₂₂The time, judge earlier and locate the information change minimum direction in pixel (2,2).Calculating is at the color-values Grad in different directions at this place, for example Grad on level and the vertical direction

$\mathrm{\ΔH}=\left|\frac{G_{21}-G_{23}}{2}\right|,$ And formula 1

$\mathrm{\ΔV}=\left|\frac{G_{12}-G_{32}}{2}\right|.$ Formula 2

If Δ H＜Δ V, think that then the information change minimum direction is a horizontal direction, therefore with the color-values interpolation green tint value G of the neighborhood pixels on the horizontal direction ₂₂, for example

G ₂₂＝(G ₂₁+G ₂₃)÷2。Formula 3

If Δ H＞Δ V, think that then the information change minimum direction is a vertical direction, therefore with the color-values interpolation green tint value G of the neighborhood pixels on the vertical direction ₂₂, for example

G ₂₂＝(G ₁₂+G ₃₂)÷2。Formula 4

This method is simple and effective when the color information of image changes extremely obviously, yet, when horizontal gradient value Δ H and vertical gradient value Δ V are close, misjudgment just takes place easily, and particularly big at noise, color information changes and has high frequency in wide lines of small, an about pixel or the image when changing (for example Nyquist pattern) regional.

Therefore, a kind of determination methods that increases the image local information change minimum direction of accuracy is to be the institute Ji.

Summary of the invention

The objective of the invention is to propose a kind of determination methods that increases the image local information change minimum direction of accuracy.

According to the present invention, a kind of method of utilizing majority rule to judge image local information change minimum direction, be included in obtain at least two image directions a plurality of in image the assessed value of the color information change direction at an interested pixel place, and adopt majority rule decision image local information change minimum direction with the represented direction of these a plurality of assessed values. the preferably, the number of these a plurality of assessed values is an odd number, and be at least three. in one embodiment, this at least two images direction comprises horizontal direction and vertical direction.

These a plurality of assessed values are the color-values generations from interested pixel and neighborhood pixels thereof, and each assessed value is represented the trend that color information changes.In one embodiment, the generation of these a plurality of assessed values comprises and utilizes a plurality of methods to calculate the Grad that is in the color-values on this at least two images direction at interested pixel; And produce these a plurality of assessed values from the difference of these a plurality of Grad.In another embodiment, these a plurality of assessed values are the flag value that are selected from a plurality of neighborhood pixels of interested pixel, and each flag value writes down the trend that its corresponding pixel place color information changes.In another embodiment, these a plurality of assessed values are included in the flag value that interested pixel is in a plurality of neighborhood pixels of the difference of Grad of the color-values on this at least two images direction and interested pixel.

Because the assessed value of utilizing a plurality of expression color informations to change is judged image local information change minimum direction with majority rule, therefore improve the accuracy of judging.

Description of drawings

Fig. 1 is 5 * 5 pel array.

Fig. 2 is the flow chart of first embodiment of the invention.

Fig. 3 is the flow chart of second embodiment of the invention.

100 pel arrays

200 judge the flow chart of image local information change minimum direction

210 calculate G _H, G _V, G _MeanWith TH0

220 judge whether to be positioned at smooth region

230 bilinearity interpolation methods

240 calculate each Grad

250 calculate the difference of all directions Grad

260 adopt majority rule with the difference of Grad judges the information change minimum direction

270 with the required color-values of method of weighting reconstruction

280 with the required color-values of method of weighting reconstruction

300 judge the flow chart of image local information change minimum direction

310 calculate the Grad on the vertical and horizontal direction

320 judge whether to have very significantly, and color information changes

Grad on 330 comparison levels and the vertical direction

340 set flag value and interpolation green tint value

350 set flag value and interpolation green tint value

360 obtain flag value Flag ₀₀, Flag ₀₂With Flag ₂₀

370 adopt majority rule with flag value judges the information change minimum direction

380 decision information change minimum direction and interpolation green tint values

390 decision information change minimum direction and interpolation green tint values

Embodiment

The present invention is that the assessed value of utilizing a plurality of expression color informations to change is judged image local information change minimum direction with majority rule, these a plurality of assessed values comprise from the color-values of interested pixel and neighborhood pixels thereof and directly producing, for example calculate the difference of the Grad of the color-values on different directions with diverse ways, or produce indirectly from the color-values of interested pixel and neighborhood pixels thereof, for example choose the flag value of a plurality of neighborhood pixels of interested pixel, each flag value writes down the trend that its corresponding pixel place color information changes.

In one embodiment, one utilizes majority rule to judge the method for image local information change minimum direction, be included at least two image directions to utilize and be at least the Grad that three odd number distinct methods calculates the color-values on this direction respectively, thereby calculate the difference of the Grad of the color-values on this at least two images direction, each difference obtains multimedia message of the same colour breath change direction, by the direction decision of the more person of occurrence number in these a plurality of color information change direction local information change minimum direction at the interested pixel place.For example select horizontal direction and vertical direction, with the Grad of the color-values of diverse ways calculating on this two direction, first method obtains Δ H ₁And Δ V ₁, second method obtains Δ H ₂And Δ V ₂, third party's method obtains Δ H ₃And Δ V ₃, calculate the difference of these a plurality of Grad

D ₁=Δ H ₁-Δ V ₁, formula 5

D ₂=Δ H ₂-Δ V ₂, and formula 6

D ₃＝ΔH ₃-ΔV ₃。Formula 7

D ₁For on the occasion of or negative value, represent that the color information change direction that first method obtains is vertical direction or horizontal direction, in like manner, D ₂And D ₃On the occasion of or negative value represent that the color information change direction that second method and third party's method obtain is vertical direction or horizontal direction.By D ₁, D ₂And D ₃Adopt majority rule and judge image local information change minimum direction.If D ₁, D ₂And D ₃In have at least two to be not less than zero, determine that then image local information change minimum direction is a vertical direction, otherwise, then be horizontal direction.

More specifically, be example with the pixel (2,2) in 5 * 5 pel arrays 100 of Fig. 1, when judging its local information change minimum direction, at first utilize the Grad of its contiguous pixel, for example with the color-values on distinct methods calculated level and the vertical direction

Δ H ₁=| G ₁₀-G ₂₃|+| G ₁₄-G ₂₁|, formula 8

Δ V ₁=| G ₀₁-G ₃₂|+| G ₁₂-G ₄₁|, formula 9

Δ H ₂=| G ₃₀-G ₂₃|+| G ₃₄-G ₂₁|, formula 10

Δ V ₂=| G ₀₃-G ₃₂|+| G ₁₂-G ₄₃|, formula 11

Δ H ₃=| G ₁₀-G ₃₄|+| G ₃₀-G ₁₄|, and formula 12

ΔV ₃＝|G ₀₁-G ₄₃|+|G ₀₃-G ₄₁|。Formula 13

In a further embodiment, the calculating of Grad comprises the function that use is made up of the color-values of the same color of the neighborhood pixels of pixel (2,2), for example f (G ₁₀, G ₁₂, G ₁₄, G ₃₀, G ₃₂, G ₃₄), f (G ₀₁, G ₂₁, G ₄₁, G ₀₃, G ₂₃, G ₄₃), or other various Grad computational methods or its combination, for example

Δ H ₁=| G ₂₁-G ₂₃|, formula 14

Δ V ₁=| G ₁₂-G ₃₂|, formula 15

Δ H ₂=| K ₁* G ₁₂-G ₁₀-G ₁₄|+| K ₂* G ₃₂-G ₃₀-G ₃₄|, formula 16

Δ V ₂=| K ₃* G ₂₁-G ₀₁-G ₄₁|+| K ₄* G ₂₃-G ₀₃-G ₄₃|, formula 17

Δ H ₃=Δ H ₁+ | G ₁₀-G ₂₃|, and formula 18

ΔV ₃＝ΔV ₁+|G ₀₁-G ₃₂|。Formula 19

Wherein, K ₁, K ₂, K ₃And K ₄Be weight.In another embodiment, the calculating of Grad comprises the different color-values of the neighborhood pixels that uses pixel (2,2), for example

Δ H ₁=| G ₂₁-G ₂₃|, formula 20

Δ V ₁=| G ₁₂-G ₃₂|, formula 21

Δ H ₂=| B ₂₂-B ₂₀|+| B ₂₂-B ₂₄|, formula 22

Δ V ₂=| B ₂₂-B ₀₂|+| B ₂₂-B ₄₂|, formula 23

Δ H ₃=| R ₁₁-R ₁₃|+| R ₃₁-R ₃₃|, and formula 24

ΔV ₃＝|R ₁₁-R ₃₁|+|R ₁₃-R ₃₃|。Formula 25

After using various account forms to obtain Grad on all directions, utilize the difference of formula 5 to 7 compute gradient values, if D ₁, D ₂And D ₃In have at least two to be not less than zero, then local information change minimum direction is a vertical direction, otherwise, then be horizontal direction.

In the above-described embodiments, on each direction all with three kinds of method compute gradient values, in other embodiments, for improving the accuracy of judging, on each direction, calculate more odd number Grad, produce the difference of more odd number Grad, each difference obtains multimedia message breath change direction of the same colour, is determined the local information change minimum direction of interested pixel at last by the more person's of these a plurality of color information change direction occurrence numbers direction.

Further, Fig. 2 shows flow process Figure 200 of a preferred embodiment, with reference to 5 * 5 pel arrays 100 of figure 1, judges the information change minimum direction of locating in pixel (2,2).At first in step 210, utilize the mean value G of neighborhood pixels on level and vertical direction of formula 26 to 29 calculating pixels (2,2) _HAnd G _V, G _HWith G _VMean value G _Mean, and by G _MeanProduce the first critical value TH0, for example

G _H=(G ₂₁+ G ₂₃)/2, formula 26

G _V=(G ₁₂+ G ₃₂)/2, formula 27

G _Mean=(G _H+ G _V)/2, and formula 28

TH0=G _Mean/ 32 or G _Mean/ 16 or G _Mean/ 8.Formula 29

Judge in step 220 whether pixel (2,2) is positioned at smooth region.If G _HWith G _VBetween absolute difference be not more than default second a critical value TH1 and the difference of the first critical value TH0, promptly | G _H-G _V|≤TH1-TH0, remarked pixel (2,2) is positioned at smooth region, thereby carry out step 230, rebuilds the green tint value G of pixel (2,2) with bilinear interpolation method ₂₂, G for example ₂₂=G _MeanOtherwise, if | G _H-G _V|＞TH1-TH0, remarked pixel (2,2) are positioned at smooth region, then adopt majority rule to judge its information change minimum direction.Therefore, in step 240, calculate Grad on all directions, for example the Δ H of formula 14 to 19 in three the different mode of odd number of being at least ₁, Δ V ₁, Δ H ₂, Δ V ₂, Δ H ₃And Δ V ₃, then at step 250 difference of a Grad, for example D of formula 5 to 7 of counting that fall into a trap ₁, D ₂And D ₃, later step 260 is adopted majority rule with these a plurality of differences and is judged its information change minimum direction.If D ₁, D ₂And D ₃In have at least two to be not less than zero, being illustrated in the information change minimum direction that pixel (2,2) locates is vertical direction, thereby carries out step 270 is rebuild pixel (2,2) with the method for weighting green tint value G ₂₂, for example

G ₂₂=W _H* G _H+ W _V* G _V, W _H＜W _VFormula 30

Otherwise being illustrated in the information change minimum direction that pixel (2,2) locates is horizontal direction, thereby carries out step 280 is rebuild pixel (2,2) with the method for weighting green tint value G ₂₂, for example

G ₂₂＝W _H×G _H+W _V×G _V，W _H＞W _V。Formula 31

In another embodiment, one utilizes majority rule to judge the method for image local information change minimum direction, the flag value that comprises a plurality of neighborhood pixels of an interested pixel in this image of statistics, each flag value writes down the trend of its corresponding pixel place information change direction, with the information change minimum direction of majority rule judgement at this interested pixel place. the method for decision flag value comprises known technology or its improvement, for example calculate the Grad of pixel color-values in different directions, the size of the Grad of the color-values on this different directions relatively, and determining the flag value of this pixel by the direction of smaller in these a plurality of Grad. Fig. 3 shows the flow chart 300 of a preferred embodiment, 5 * 5 pel arrays 100 with reference to figure 1, judgement is in pixel (2,2) the information change minimum direction of locating. at first calculate the level located in pixel (2,2) and the Grad on the vertical direction in step 310

$\mathrm{\ΔH}=\left|\frac{G_{21}-G_{23}}{2}\right|,$ And formula 32

$\mathrm{\ΔV}=\left|\frac{G_{12}-G_{32}}{2}\right|.$ Formula 33

In step 320, judge in pixel (2,2) and locate whether to have extremely significantly color information variation.If the absolute difference of Δ H and Δ V is greater than a critical value TH1 who sets, promptly | Δ H-Δ V|＞TH1 is illustrated in pixel (2,2) and locates to have extremely significantly color information variation, thereby carries out color information change direction and the flag value thereof of step 330 decision at this place.In step 330, if the Grad of horizontal direction is less than the Grad of vertical direction, i.e. Δ H＜Δ V, being illustrated in the color information change direction that pixel (2,2) locates is horizontal direction, therefore sets its flag value Flag in step 340 ₂₂Be horizontal direction, and with the green tint value G of the color-values interpolated pixel (2,2) of pixel (2,2) neighborhood pixels in the horizontal direction ₂₂, for example

G ₂₂=(G ₂₁+ G ₂₃)/2; Formula 34

Otherwise being illustrated in the color information change direction that pixel (2,2) locates is vertical direction, therefore sets its flag value Flag in step 350 ₂₂Be vertical direction, and with the green tint value G of the color-values interpolated pixel (2,2) of pixel (2,2) neighborhood pixels in vertical direction ₂₂, for example

G ₂₂＝(G ₁₂+G ₃₂)/2。Formula 35

Be not more than critical value TH1 if before in step 320, obtain the absolute difference of Δ H and Δ V, be illustrated in pixel (2,2) locating not have extremely significantly, color information changes, therefore in step 360, obtain the flag value of a plurality of neighborhood pixels of pixel (2,2), for example select pixel (0,0), (0,2) with (2,0), its flag value is respectively Flag ₀₀, Flag ₀₂With Flag ₂₀Flag value can also can directly be obtained through the pixel of judging by previous by the relatively size generation of Grad.Then carry out step 370, with the flag value Flag of these a plurality of neighborhood pixels ₀₀, Flag ₀₂With Flag ₂₀Adopt majority rule and judge local information change minimum direction.If Flag ₀₀, Flag ₀₂With Flag ₂₀In have at least two for horizontal direction, being illustrated in the information change minimum direction that pixel (2,2) locates is horizontal direction, therefore in step 380 with the green tint value G of the color-values interpolated pixel (2,2) of pixel (2,2) neighborhood pixels in the horizontal direction ₂₂, for example formula 34; Otherwise being illustrated in the information change minimum direction that pixel (2,2) locates is vertical direction, therefore in step 390 with the green tint value G of the color-values interpolated pixel (2,2) of pixel (2,2) neighborhood pixels in vertical direction ₂₂, for example formula 35.

Characteristics of the present invention are to utilize majority rule to judge image local information change minimum direction.In the embodiment of Fig. 2, be that the difference of the Grad that calculates by diverse ways is carried out majority rule and judged.In the embodiments of figure 3, be to carry out majority rule by the flag value of a plurality of neighborhood pixels to judge.In other embodiments, adopt the difference of a plurality of Grad and a plurality of flag value to carry out the majority rule judgement simultaneously.Judge for carrying out majority rule, the preferably, the difference of Grad or the number of flag value are odd number, and are at least three.Improving its number can increase the accuracy of judgement.

The foregoing description is only in order to explanation the present invention, and non-limiting the present invention.

Claims (20)

1. a method of utilizing majority rule to judge image local information change minimum direction is characterized in that, comprises the following steps:

On vertical and horizontal both direction, obtain respectively equal number and be at least Grad three, odd number color information change direction at pixels of interest place in image or flag value as assessed value; And

Adopt majority rule fixing local information change minimum direction with the represented direction of described a plurality of assessed values.

2. the method for claim 1 is characterized in that, the step that obtains described a plurality of assessed values comprises the following steps:

Utilize a plurality of methods to calculate the Grad that is in the color-values on the described vertical and horizontal both direction at interested pixel; And

Produce described a plurality of assessed value from the difference of described a plurality of Grad.

3. method as claimed in claim 2 is characterized in that, the step of calculating described a plurality of Grad comprises the Grad of color-values of the same color of a plurality of neighborhood pixels that calculate interested pixel.

4. method as claimed in claim 2 is characterized in that, the step of calculating described a plurality of Grad comprises the Grad of color-values of a plurality of colors of a plurality of neighborhood pixels that calculate interested pixel.

5. the method for claim 1 is characterized in that, described a plurality of assessed values comprise the flag value of a plurality of neighborhood pixels that are selected from interested pixel, and each flag value writes down the trend that its corresponding pixel place color information changes.

6. method as claimed in claim 5 is characterized in that the decision of each flag value comprises the following steps:

Calculating is in the Grad of the color-values on the described vertical and horizontal both direction in its corresponding pixel;

Judge from described a plurality of Grad whether the pixel in correspondence has extremely significantly color information variation; And

Have when extremely significantly color information changes at the pixel place of correspondence, more described a plurality of Grad size each other is with the decision flag value.

7. the method for claim 1 is characterized in that, the step that obtains described a plurality of assessed values comprises the following steps:

Utilize a plurality of methods to calculate the Grad that is in the color-values on the described vertical and horizontal both direction at interested pixel;

Produce a plurality of first assessed values from the difference of described a plurality of Grad;

The flag value of a plurality of neighborhood pixels of selection interested pixel is as second assessed value, and each flag value writes down the trend that its corresponding pixel place color information changes; And

Use one of first and second assessed value as described a plurality of assessed values.

8. method as claimed in claim 7 is characterized in that, the step of calculating described a plurality of Grad comprises the Grad of color-values of the same color of a plurality of neighborhood pixels that calculate interested pixel.

9. method as claimed in claim 7 is characterized in that, the step of calculating described a plurality of Grad comprises the Grad of color-values of a plurality of colors of a plurality of neighborhood pixels that calculate interested pixel.

10. method as claimed in claim 7 is characterized in that the decision of each flag value comprises the following steps:

Calculating is in the Grad of the color-values on the described vertical and horizontal both direction in its corresponding pixel;

Judge from described a plurality of Grad whether the pixel in correspondence has extremely significantly color information variation; And

Have when extremely significantly color information changes at the pixel place of correspondence, more described a plurality of Grad size each other is with the decision flag value.

11. a method of utilizing majority rule to judge image local information change minimum direction is characterized in that, comprises the following steps:

Judge whether an interested pixel is positioned at a smooth region in the image, carry out subsequent step when being positioned at a smooth region when this interested pixel is non-;

On vertical and horizontal both direction, obtain respectively equal number and be at least Grad three, odd number color information change direction at pixels of interest place in image or flag value as assessed value; And

Adopt majority rule fixing local information change minimum direction with the represented direction of described a plurality of assessed values.

12. method as claimed in claim 11 is characterized in that, the step that obtains described a plurality of assessed values comprises the following steps:

Utilize a plurality of methods to calculate the Grad that is in the color-values on the described vertical and horizontal both direction at interested pixel; And

Produce described a plurality of assessed value from the difference of described a plurality of Grad.

13. method as claimed in claim 12 is characterized in that, the step of calculating described a plurality of Grad comprises the Grad of color-values of the same color of a plurality of neighborhood pixels that calculate interested pixel.

14. method as claimed in claim 12 is characterized in that, the step of calculating described a plurality of Grad comprises the Grad of color-values of a plurality of colors of a plurality of neighborhood pixels that calculate interested pixel.

15. method as claimed in claim 11 is characterized in that, described a plurality of assessed values comprise the flag value of a plurality of neighborhood pixels that are selected from interested pixel, and each flag value writes down the trend that its corresponding pixel place color information changes.

16. method as claimed in claim 15 is characterized in that, the decision of described each flag value comprises the following steps:

Calculating is in the Grad of the color-values on the described vertical and horizontal both direction in its corresponding pixel;

Judge from described a plurality of Grad whether the pixel in correspondence has extremely significantly color information variation; And

Have when extremely significantly color information changes at the pixel place of correspondence, more described a plurality of Grad size each other is with the decision flag value.

17. method as claimed in claim 11 is characterized in that, the step that obtains described a plurality of assessed values comprises the following steps:

Utilize a plurality of methods to calculate the Grad that is in the color-values on the described vertical and horizontal both direction at interested pixel;

Produce a plurality of first assessed values from the difference of described a plurality of Grad;

The flag value of a plurality of neighborhood pixels of selection interested pixel is as second assessed value, and each flag value writes down the trend that its corresponding pixel place color information changes; And

Use one of first and second assessed value as described a plurality of assessed values.

18. method as claimed in claim 17 is characterized in that, the step of calculating described a plurality of Grad comprises the Grad of color-values of the same color of a plurality of neighborhood pixels that calculate interested pixel.

19. method as claimed in claim 17 is characterized in that, the step of calculating described a plurality of Grad comprises the Grad of color-values of a plurality of colors of a plurality of neighborhood pixels that calculate interested pixel.

20. the method for claim 1 is characterized in that, the decision of each flag value comprises the following steps:

Calculating is in the Grad of the color-values on the described vertical and horizontal both direction in its corresponding pixel;

From this a plurality of Grad judge whether have pixel in correspondence extremely significantly color information changes; And

Have when extremely significantly color information changes at the pixel place of correspondence, more described a plurality of Grad size each other is with the decision flag value.

Images