CN105260986B - A kind of image magnification method of anti - Google Patents

Abstract

The invention discloses a kind of image magnification methods of anti, the evaluation index of image fog-level is obtained according to the gradient of input picture, in conjunction with amplification factor, image resolution ratio, ask for image sharpening intensive parameter, under the control of the parameter, the sharpening being adapted with quality of input image grade is performed；And then the overall amplification process of image is decomposed into the amplification step by step of several small multiplying powers, and it is 1.2 times high using the common each enlarged drawing image width of interpolation method, until meeting overall magnification requirement.The present invention effectively eliminates the intrinsic blurring effect of interpolation method, significantly improves the clarity of enlarged drawing under the premise of unobvious increase computational complexity.

Description

An Anti-blurring Image Enlargement Method

technical field

The invention belongs to the technical field of image processing, and relates to an image interpolation enlarging method, in particular to an anti-blurring image enlarging method.

technical background

Currently, digital cameras or cameras can easily capture high-resolution videos, but for the sake of saving storage or transmission costs, there are still a large amount of low-resolution video content in applications such as video surveillance, video conferencing, and Internet TV. Video upscaling technology is therefore widely used to improve the spatial resolution of images to meet the requirements of the actual display area of the screen. Some special applications, such as video surveillance, also involve zooming in on a local area of the image so that details of the image can be identified.

A large number of video image scaling techniques have been developed, ranging from simple interpolation to complex image super-resolution. Although image super-resolution technology can obtain a clearer zoom-in effect, the computational complexity remains high; what is more inconvenient is that image super-resolution generally requires training sample libraries in advance, which lacks universality and scalability in practical applications. operability. Therefore, the current image enlargement technology used in video application software or media players is mainly based on simple and fast interpolation methods, such as bilinear interpolation (Bilinear), bicubic interpolation (Bicubic), Lanczos interpolation (Lanczos) Wait. However, these classic interpolation and amplification methods also have a serious defect. When the original image is not clear (such as a compressed image) or the magnification is too large, the image after interpolation and amplification has obvious blurring effect, which affects viewing. the subjective experience of the reader.

The spatial sharpening filter is a high-pass filter that can strengthen the high-frequency components of the digital image, just opposite to the low-pass effect of the interpolation filter. Therefore, if a sharpening operation is inserted before image interpolation and magnification, it will be possible to offset the blurring effect caused by interpolation to a certain extent. In addition, the higher the single magnification of the image, the easier it is to cause blur, on the contrary, the lower the magnification, the lighter the blur. Therefore, if the high magnification process is converted into several low magnification steps, it will help to improve the quality of the magnified image. Based on these two angles, the present invention proposes an anti-blurring image enlargement method.

Contents of the invention

In order to solve the above technical problems, the present invention cascades the image sharpening operation and the interpolation operation into a whole, enhances the contour details of the image through sharpening preprocessing, and contributes the signal source of high-frequency component enhancement to the interpolation operation, thereby eliminating the interpolation process. Low pass blur effect effect.

The technical solution adopted in the present invention is: a kind of anti-blurring image enlargement method, it is characterized in that, comprises the following steps:

Step 1: Use the gradient feature of the image to measure the blur degree of the image, and calculate the blur index of the input image;

Step 2: Calculate the sharpening intensity parameter of the input image by integrating the three elements of the blur index, the overall magnification, and the image resolution of the input image;

Step 3: Call the sharpening algorithm to sharpen the input image according to the calculated sharpening intensity constant;

Step 4: Decompose the enlargement operation with the overall enlargement factor of R into a number of progressive enlargements with small magnifications, and determine the execution times n of successive enlargements;

Step 5: Perform n times of step-by-step interpolation and amplification operations.

Preferably, the calculation of the ambiguity index described in step 1 adopts the following method:

Define the ambiguity index Fb as the energy-normalized gradient:

Among them, Gx and Gy respectively represent the horizontal and vertical gradient maps, obtained by Sobel, Roberts or Prewitt gradient operators; E is the energy of the input image, calculated with the input image pixel I ".*" means matrix dot product operation.

As a preference, the calculation of the sharpening intensity parameter λ described in step 2 adopts the following method:

Among them, Fb is the blur index, R is the overall magnification, S is the area of the input image, which is equal to the length multiplied by the height, and _Scif is a constant based on the area of the CIF format image.

Preferably, the sharpening operation described in step 3 is to superimpose the high-pass filtered version of the input image on the original input image after adjustment, and the calculation formula is:

y(n,m)=x(n,m)+λz(n,m);

Here, x and y represent the original input image and the sharpened input image respectively, z represents the input image after high-pass filtering, which is obtained by a two-dimensional Laplacian differential operator, and λ is the sharpening intensity parameter.

As a preference, the determination of the execution times n of step-by-step amplification described in step 4 adopts the following rules:

Each time the width and height of the image are enlarged by 1.2 times, if the overall magnification factor is R, then the execution times n of enlarging 1.2 times each time and enlarging to R times successively is calculated as symbol Indicates rounding down.

Preferably, in step 5, performing n times of step-by-step interpolation and amplification operations, the interpolation methods include but not limited to bilinear interpolation, bicubic interpolation, and Lanz interpolation.

Compared with the existing image enlargement scheme, the present invention has the following advantages and positive effects:

(1) The present invention applies a sharpening pretreatment in front of the simple interpolation method, effectively eliminates the inherent fuzzy effect of the interpolation method, and significantly improves the clarity of the enlarged image without significantly increasing the computational complexity;

(2) Compared with complex super-resolution schemes, the method of the present invention does not involve sample image training, and has the advantages of simple implementation, high processing efficiency, and strong universality;

(3) Both the sharpening algorithm and the interpolation algorithm based on the method of the present invention come from mature technologies, which are convenient for assembling existing modules for integration and realization.

Description of drawings

Fig. 1: processing flowchart of the embodiment of the present invention;

Fig. 2: the comparison example diagram of the magnification effect of the embodiment of the present invention, wherein (a) is the original image enlarged by 4 times through the ordinary bilinear effect diagram, (b) is the original image enlarged by the 4 times effect diagram by the method of the present invention, ( c) is the 4-fold effect diagram of the decoded image enlarged by ordinary bilinear, and (d) is the 4-fold effect diagram of the decoded image enlarged by the method of the present invention.

Detailed ways

In order to facilitate those of ordinary skill in the art to understand and implement the present invention, the present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the implementation examples described here are only used to illustrate and explain the present invention, and are not intended to limit this invention.

The invention provides an anti-blurring image enlargement method, which cascades image sharpening operations and interpolation operations into a whole. The intensity of the sharpening operation should be adapted to the quality of the input image itself. Clear images should be weakly sharpened, and blurred images should be strongly sharpened. Otherwise, sharpening will not only fail to provide preprocessed images with enhanced high-frequency components for subsequent interpolation operations , which will damage the image and reduce visual fidelity. The factors that cause image blur come from many aspects, such as motion blur caused by the relative motion of the target and the camera, lens defocus, atmospheric scattering, image compression, noise reduction and resampling and other filtering processes, and so on. These processes without exception cause the loss of image details and weaken the gradient of the target edge contour. Therefore, the gradient feature of the image is used to describe the blur degree of the image. The gradient map of the clear image is sharp, and the gradient map of the blurred image is flat.

Image sharpening algorithms usually use a sharpening strength parameter to control the degree of sharpening, and at least three factors should be considered for the degree of sharpening applied to the input image. The first and most important is the degree of blur of the image, which can be measured by the blur index; the second is the magnification of the image. The higher the magnification, the easier it is to cause the blur effect of low-pass interpolation. Accordingly, stronger sharpening should be performed to compensate for the magnification The blurring effect; the last is the resolution of the input image itself. The lower the resolution of the image, the worse the detail retention, and the more it needs to be sharpened to enhance the details.

Based on the above considerations, the complete processing flow of the method of the present invention is shown in Figure 1, comprising the following steps:

Step 1: Use the gradient feature of the image to measure the blur degree of the image, and calculate the blur index of the input image;

Integrating the influence of image intensity, the blur index Fb is defined as the energy normalized gradient

Here, Gx and Gy respectively represent the horizontal and vertical gradient maps, which can be obtained by classic gradient operators such as Sobel, Roberts, Prewitt, etc., E is the energy of the input image, and is calculated using the input image pixel I ".*" means matrix dot product operation.

Step 2: Calculate the sharpening intensity parameter of the input image by integrating the three elements of the blur index, the overall magnification, and the image resolution of the input image;

Combined with empirical data, the calculation of the sharpening intensity parameter λ adopts the following method:

Among them, Fb is the blur index, R is the overall magnification, S is the area of the input image, which is equal to the length multiplied by the height, and S _cif is a constant based on the area of the CIF format image (352x288).

Step 3: Call the sharpening algorithm to sharpen the input image according to the calculated sharpening intensity constant;

The sharpening algorithm used in the present invention is to superimpose the high-pass filter version of the input image on the original input image after adjustment, and the calculation formula is

y(n,m)=x(n,m)+λz(n,m);

Here, x and y represent the original input image and the sharpened input image respectively, z represents the high-pass filtered input image, which is obtained by a two-dimensional Laplacian differential operator, and λ is the sharpening intensity parameter.

Step 4: Decompose the enlargement operation with the overall enlargement factor of R into a number of progressive enlargements with small magnifications, and determine the execution times n of successive enlargements;

Each time the width and height of the image are enlarged by 1.2 times, if the overall magnification factor is R, then the execution times n of enlarging 1.2 times each time and enlarging to R times successively is calculated as symbol Indicates rounding down.

Step 5: Perform n times of step-by-step interpolation operations, and the interpolation methods available for selection include but are not limited to bilinear interpolation, bicubic interpolation, and Ranz interpolation.

It should be understood that the parts not described in detail in this specification belong to the prior art.

It should be understood that the above-mentioned descriptions for the preferred embodiments are relatively detailed, and should not therefore be considered as limiting the scope of the patent protection of the present invention. Within the scope of protection, replacements or modifications can also be made, all of which fall within the protection scope of the present invention, and the scope of protection of the present invention should be based on the appended claims.

Claims (5)

1. a kind of image magnification method of anti, which is characterized in that include the following steps：

Step 1：Image fog-level, the fuzziness index of calculating input image are weighed using the Gradient Features of image；

The calculating of the fuzziness index, with the following method：

Fuzziness index Fb is defined as to the gradient of energy normalized：

Wherein, Gx, Gy represent horizontal and vertical gradient figure respectively, are asked by Sobel, Roberts or Prewitt gradient operator It takes；E is the energy of input picture, is calculated with input image pixels I" .* " representing matrix dot-product operation；

Step 2：Three elements of fuzziness index, overall magnification, image resolution ratio of comprehensive input picture calculate input figure The sharpening intensities parameter lambda of picture；

Step 3：According to the sharpening intensities parameter lambda being calculated, sharpening algorithm is called to be sharpened input picture；

Step 4：The amplifieroperation that overall magnification is R is decomposed into the amplification step by step of several small multiplying powers, determines to put step by step Big execution frequency n；

Step 5：Perform n times interpolation amplification operation step by step.

2. the image magnification method of anti according to claim 1, it is characterised in that：Sharpening described in step 2 is strong The calculating of parameter lambda is spent, with the following method：

Wherein, Fb is fuzziness index, and R is overall magnification, and S is the area of input picture, multiplies height, S equal to length_cifIt is one Constant on the basis of CIF format-pattern areas.

3. the image magnification method of anti according to claim 1, it is characterised in that：Calling described in step 3 is sharp Change algorithm to be sharpened input picture, be that the high-pass filtered version of input picture is added to be originally inputted figure after adjustment As upper, calculation formula is：

Y (n, m)=x (n, m)+λ z (n, m)；

Here, x, y represent the input picture after original input picture and sharpening respectively, and z represents the input picture after high-pass filtering, It is obtained with two-dimentional Laplacian differential operators, λ is sharpening intensities parameter.

4. the image magnification method of anti according to claim 1, it is characterised in that：Described in step 4 determine by The execution frequency n of grade amplification, using following rule：

Every time by the width of image and 1.2 times highly enlarged, if overall magnification is R, every time 1.2 times of amplification, gradually put Big to R times of execution frequency n is calculated asSymbolRepresent lower rounding.

5. the image magnification method of anti according to claim 1, it is characterised in that：Execution n times described in step 5 Interpolation amplification operation step by step, interpolation method include but not limited to bilinear interpolation, bi-cubic interpolation method, blue hereby interpolation method.