CN1804901A - Neighbor diffused selection based smooth image multiplication method - Google Patents

Abstract

The invention relates to a neighborhood spread selection smooth image amplification algorism in the field of an image processing. It comprises three steps: 1) dispersing the image. 2) Using the neighborhood spread method to obtain the amplified image f0. 3) Applying the spread wave filter to do smooth and rim reinforce to f0. It is mainly used in image processing.

Description

Neighbor diffused selection based smooth image multiplication method

Technical field

The present invention relates to image processing field, specifically relate to neighbor diffused selection based smooth image multiplication method.

Background technology

The amplifying technique of digital picture all plays an important role in a lot of fields such as medical science, military affairs, meteorology.Yet, traditional image multiplication method such as arest neighbors interpolation, methods such as bilinear interpolation and bicubic interpolation all exist place not fully up to expectations on treatment effect, as blocking effect, the sawtooth effect at edge, or carry out that the border causes smooth the time image blurring or the like.These shortcomings have caused all inconvenience in the practical application, therefore, how to avoid the generation of above-mentioned shortcoming effectively, and obtaining high-quality enlarged image has become research focus in the image processing field.

Diffusing filter is a kind of non-linear filtering method that grows up the nearly more than ten years, it has broken the constraint of original linear filter, can in smoothed image, keep edge of image preferably, thereby obtain paying close attention to widely in image processing field based on diffusion algorithm.Aspect the image amplification, 2003, S.Battiato, G.Gallo and F.Stanco first with the nonlinear diffusion method as a part of image multiplication method and introduce into and obtain the SIAD algorithm, they at first utilize the bicubic interpolation method image to be amplified to the multiple of requirement, the edge enhancing method that utilizes G.Leu to propose in 2000 again carries out the edge to it and strengthens, and removes the culture noise of introducing with low-pass filter at last.Henceforth, the researchist begins to be devoted to seek more suitable partial differential equation, focuses on more in the said method improvement in second step, and has ignored the importance of the first step in whole algorithm.Well imagine,, then raise the efficiency and strengthen to amplify and to obtain the effect of getting twice the result with half the effort aspect the quality in whole algorithm if used simple in a first step or amplify the measured method of matter.

Summary of the invention

For overcoming the deficiencies in the prior art, the object of the present invention is to provide a kind ofly to be easy to realize simple and effect good image multiplication method.

The technical solution used in the present invention is:

The present invention includes following three steps:

(1) thinization of image

Thinization of image is that zero row and zero row that insert some between row and column adjacent in original image are respectively realized, image size after thinization is big or small consistent with amplification back image, if f is (x, y) digital picture of the capable N row of expression M size, x wherein, y represents the coordinate of row and column respectively, and desire is amplified K doubly with it, then its thinization image f _s(x y) is expressed as:

(2) utilize the neighborhood diffusion method to obtain enlarged image f ₀

The thinization image of need enlarged image can be obtained enlarged image with corresponding neighborhood diffusion filter convolution, two coefficient filter that the neighborhood diffusion filter is made up of " 1 " and " 0 ", its impulse response is defined as follows: when enlargement factor K is odd number:

When K is an even number, x, the y value is ± 1/2, ± 3/2, ± 5/2 ... the time:

Then (x, enlarged image y) is: f corresponding to f in the step (1) ₀(x, y)=f _s(x, y) * h _oOr f ₀(x, y)=f _s(x, y) * h _eWherein " * " represents convolution;

(3) use diffusion filter to f ₀Carry out level and smooth and the edge enhancing

Use diffusion filter to f ₀Carry out level and smooth and the edge enhancing, selective filter is the selection smoothing filter of Catte, and its equation is

$\frac{\∂u}{\∂t}=\mathrm{div}[g\left(\right|G_{\mathrm{\σ}}*\▿u\left|\right)\▿u]$

Div represents the content in the square bracket is asked its divergence in the formula; G () is the diffusivity function, can choose arbitrarily as required, and generally requiring it is subtraction function; $G_{\mathrm{\σ}}=\frac{1}{2{\mathrm{\π\σ}}^2}\exp (-(x^2+y^2)/{2\mathrm{\σ}}^2)$ Be Gaussian function, wherein σ is a standard deviation, be one greater than zero parameter, can set as required; " || " expression delivery, the gradient of " u " expression u; The starting condition of equation is u ₀(x, y)=f ₀(x, y), separating of equation is the enlarged image that this algorithm finally obtains.

The present invention possesses following effect: the present invention is by using a kind of simple interpolation method one neighborhood diffusion method that is complementary with the human visual system, combine with the Catte diffusion filter, obtained a kind ofly being easy to realize, simple and the good image multiplication method one neighbor diffused selection based smooth method of effect (adjacent diffusion and selective smoothing, ADASS).

Description of drawings

Fig. 1: neighborhood diffusion pattern

Fig. 2: utilize the ADASS algorithm to amplify 4 times image

Fig. 3: utilize algorithms of different image to be amplified 4 times result

Fig. 4: algorithm flow chart of the present invention

Embodiment

Further specify the present invention below in conjunction with drawings and Examples.

The present invention has used a kind of simple interpolation method one neighborhood diffusion method that is complementary with the human visual system, combine with the Catte diffusion filter, obtained a kind ofly being easy to realize, simple and the good image multiplication method of effect-neighbor diffused selection based smooth method (adjacent diffusion and selective smoothing, ADASS), algorithm of the present invention comprises following three steps:

1. thinization of image

It is to utilize the first step of method of interpolation enlarged image that image is carried out thinization of zero padding always, thinization of image is that zero row and zero row that insert some between row and column adjacent in original image are respectively realized, image is big or small consistent after the image size after thinization and the amplification.If f (x, the y) digital picture of the capable N of expression M row size, x wherein, y represents the coordinate of row and column respectively, desire is amplified K doubly with it, then its thinization image f _s(x y) is expressed as:

2. utilize neighborhood diffusion method enlarged image

Utilizing neighborhood diffusion method enlarged image is with the thinization image of need enlarged image and corresponding neighborhood diffusion filter convolution.Two coefficient filter that the neighborhood diffusion filter is made up of " 1 " and " 0 ", its impulse response is defined as follows:

When enlargement factor K is odd number:

When K is an even number, x, the y value is ± 1/2, ± 3/2, ± 5/2 ... the time:

Then (x, enlarged image y) is: f corresponding to f in the step (1) ₀(x, y)=f _s(x, y) * h _oOr f ₀(x, y)=f _s(x, y) * h _eWherein " * " represents convolution.

The frequency response characteristic of this wave filter is complementary with human visual direction selectivity, so that the method is not only compared the bicubic interpolation method is easy to be many, amplifies also not a halfpenny the worse qualitatively at image.This also is the main cause that this algorithm is selected it.The neighborhood diffusion method is interpenetrating of seed sampling point and mixes that when enlargement factor K was odd and even number, the seed sampling point was respectively with rhombus and the diffusion of accurate rhombus ring-type, as shown in Figure 1.

It is to be noted, with the convolution of neighborhood diffusion filter and be different from convolution process under other situation, need do complicated weighted sum computing to the gray scale of image, and can realize by the well-regulated conversion of simple sampling point coordinate that just this characteristic makes the neighborhood diffusion filter be particularly advantageous in hardware and realizes.

3. selection smothing filtering

The enlarged image that obtains through above-mentioned two steps is designated as f ₀, use diffusion filter in this step to f ₀Carry out level and smooth and the edge enhancing.The wave filter of Xuan Zeing is the selection smoothing filter of Catte herein, and its equation is

$\frac{\∂u}{\∂t}\mathrm{div}[g\left(\right|G_{\mathrm{\σ}}*\▿u\left|\right)\▿u]$

Div represents the content in the square bracket is asked its divergence in the formula; G () is the diffusivity function, can choose arbitrarily as required, and generally requiring it is subtraction function; $G_{\mathrm{\σ}}=\frac{1}{2{\mathrm{\π\σ}}^2}\exp (-(x^2+y^2)/{2\mathrm{\σ}}^2)$ Be Gaussian function, wherein σ is a standard deviation, be one greater than zero parameter, can set as required; " || " expression delivery, the gradient of " u " expression u; The starting condition of equation is u ₀(x, y)=f ₀(x, y).Separating of equation is the enlarged image that this algorithm finally obtains.

Because the Catte diffusion equation itself has the ability of removing noise, so neighbor diffused selection based smooth image multiplication method has saved the step of low-pass filtering.

Further specify the present invention below in conjunction with an instantiation.

1) the lena image of selection 256 * 256 and 256 * 256 I34 image experimentize to this algorithm, pick and place big multiple K=4, and then the impulse response of neighborhood diffusion filter is

$h=\left[\begin{array}{cccccc}0& 0& 1& 1& 0& 0\\ 0& 1& 0& 0& 1& 0\\ 1& 0& 1& 1& 0& 1\\ 1& 0& 1& 1& 0& 1\\ 0& 1& 0& 0& 1& 0\\ 0& 0& 1& 1& 0& 0\end{array}\right]$

With the thinization image and the h convolution of lena image and I34 image, obtain enlarged image respectively, use f _OLAnd f _OIExpression.Next the selection smoothing filter that adopts Catte is to f _OLAnd f _OICarry out level and smooth and the edge enhancing.Here, the diffusivity function is chosen as g (s)=1/ (1+ (s/ λ) ²), get σ=0.7, λ=30, iteration step length is 0.1, iteration 5 times, obtain respectively among Fig. 2 (b) and (d) figure.

2) for the ease of comparing, adopt a part (70 to 169 row of the neighbor diffused selection based smooth image multiplication method of arest neighbors interpolation, bilinear interpolation, SIAD method and this paper respectively to lena figure, 100 to 179 row) and the part of I34 image (160 to 229 row, 10 to 49 are listed as) amplify.When using the ADASS algorithm, select enlargement factor K=4 equally, g (s)=1/ (1+ (s/ λ ²), σ=0.7, λ=30, iteration step length is 0.1, iteration 5 times, the gained image is shown in (d) among Fig. 3.The amplification result of other algorithm is respectively shown in the (a) and (b) among Fig. 3, (c).

Claims (1)

1. a neighbor diffused selection based smooth image multiplication method is characterized in that, comprises following three steps:

(1) thinization of image

Thinization of image is that zero row and zero row that insert some between row and column adjacent in original image are respectively realized, image size after thinization is big or small consistent with amplification back image, if f is (x, y) digital picture of the capable N row of expression M size, x wherein, y represents the coordinate of row and column respectively, and desire is amplified K doubly with it, then its thinization image f _s(x y) is expressed as:

(2) utilize the neighborhood diffusion method to obtain enlarged image f ₀

The thinization image of need enlarged image can be obtained enlarged image with corresponding neighborhood diffusion filter convolution, two coefficient filter that the neighborhood diffusion filter is made up of " 1 " and " 0 ", its impulse response is defined as follows:

When enlargement factor K is odd number:

When K is an even number, x, the y value is ± 1/2, ± 3/2, ± 5/2 ... the time:

Then (x, enlarged image y) is: f corresponding to f in the step (1) ₀(x, y)=f _s(x, y) * h ₀Or f ₀(x, y)=f _s(x, y) * h _eWherein " * " represents convolution;

(3) use diffusion filter to f ₀Carry out level and smooth and the edge enhancing

Use diffusion filter to f ₀Carry out level and smooth and the edge enhancing, selective filter is the selection smoothing filter of Catte, and its equation is:

$\frac{\∂u}{\∂t}=\mathrm{div}[g\left(\right|G_{\mathrm{\σ}}*\▿u\left|\right)\▿u]$

Div represents the content in the square bracket is asked its divergence in the formula; G () is the diffusivity function, can choose arbitrarily as required, and generally requiring it is subtraction function; $G_{\mathrm{\σ}}=\frac{1}{2{\mathrm{\π\σ}}^2}\exp (-(x^2+y^2)/2{\mathrm{\σ}}^2)$ Be Gaussian function, wherein σ is a standard deviation, be one greater than zero parameter, can set as required; " | | " the expression delivery, the gradient of " u " expression u; The starting condition of equation is u ₀(x, y)=f ₀(x, y), separating of equation is the enlarged image that this algorithm finally obtains.

Images