CN101329762A - Fidelity Evaluation Method Based on Context-Dependent Image Resizing - Google Patents

Abstract

The fidelity evaluation method based on content-dependent image size adjustment belongs to the image fidelity evaluation technology, and is characterized in that it comprises the following steps: A. utilizing the energy function to be evaluated, and carrying out the content of the pre-selected image through the seam carving algorithm Relevant size scaling; B. Find the corresponding points of the original image in the adjusted image; C. Compare the adjusted image with the original image, and obtain the fidelity evaluation result of the energy function according to the comparison results; also disclosed An image fidelity evaluation system with different energy functions in the seam carving algorithm is provided, including: a reference image database, a seam caving algorithm unit and a fidelity evaluation unit; Fidelity evaluation, thereby realizing the image fidelity evaluation after the content-related size of the image is changed under different energy functions.

Description

Fidelity Evaluation Method Based on Context-Dependent Image Resizing

technical field

The invention relates to image fidelity evaluation technology, in particular to an image fidelity evaluation method based on content-dependent size adjustment technology.

Background technique

In modern society, we are exposed to more and more display screens. There is a wide range of sizes from small ones such as camera screens and mobile phone screens to large ones such as computer monitors and movie screens, and the aspect ratios are also different. Therefore, it is required that the same image can be changed into different sizes and displayed on screens with different sizes and different aspect ratios. Existing technologies use simple image stretching or cropping. For screens with the same aspect ratio, using stretching is still good, but for screens with different aspect ratios, it will cause image subject content to be deformed or lose information. Today, when digital TV and smart phones are gradually popularized, such a result is unsatisfactory. The content-dependent image resizing technology is a technology that uses computer algorithms to process images so that the subject content is basically unchanged. While the hardware system is becoming more and more perfect, we can clearly see the broad development prospects of content-dependent resizing technology.

In content-related image resizing technology, the seam carving algorithm is often used. [See Shai Avidan, Ariel Shamir, Seam Carving for Content-Aware ImageResizing ACM Transactions on Graphics, Volume 26, Number 3, (SIGGRAPH 2007) for the relevant content and specific implementation and operation of the seam carving algorithm] Simply put, the seam carving algorithm is to try to Remove the points with lower energy without changing the continuity of the image to adapt to the target size. The definition of energy is to assign higher energy values to the points that are greatly different from the surrounding colors, because the human eye is more recognizable to them, and at the same time, we can assign higher energy values to the content we want to leave, to meet the needs of users.

It can be seen that the seam carving algorithm is based on the energy function unit, which is what we call the energy function. It can be said that the energy function is the core and soul of the seam carving algorithm. Different energy functions will produce different seams, so the adjustment map formed by the seamcarving algorithm will also be different.

The most common energy function is the gradient function of the image. Its definition is that the energy value of each point is the absolute value of the gradient value of the point in the horizontal direction plus the absolute value of the gradient value in the vertical direction; more commonly used are other Edge detection operators, such as Prewitt operator, Sobel operator, etc.

At present, people use direct observation with human eyes to compare the pros and cons of various energy function methods. However, such psychophysical experiments obviously have disadvantages such as tedious, expensive, long cycle, difficult to automate and difficult to replicate. At present, new energy function methods are emerging in an endless stream. It is unrealistic to quickly conduct a psychophysical experiment every time a new energy function appears, and it is obviously the most ideal way to use computers to simulate intelligence evaluation.

Invention content:

In view of this, a main purpose of the present invention is to provide a method for evaluating image fidelity, which can improve the accuracy and efficiency of evaluating different energy functions in the seamcarving algorithm.

Another main purpose of the present invention is to provide an image fidelity evaluation system, which can improve the accuracy and efficiency of evaluating different energy functions in the seam carving algorithm.

The evaluation method described is as follows:

Step (1): Initialize, set up the following units in the computer to build an evaluation system:

The reference image database unit stores reference image data to be processed, including: original image, energy image, and fringe seam image;

The strip cutting algorithm unit, after inputting the energy function to be measured, performs content-related image size adjustment on the loaded reference image through the strip cutting seam carving algorithm, and records the position of each point in the original image during the adjustment process. The position of the corresponding point on the rear map obtained after the adjustment is completed, wherein there are: an energy calculation unit, a fringe search unit, an image size adjustment unit and a corresponding point calculation unit, wherein:

The energy calculation unit loads the energy function to be evaluated and loads the original image from the reference image database, calculates the energy of points in the original image to generate an energy map, and then loads the energy map into the reference image database;

The fringe search unit loads the energy map from the energy calculation unit, uses the fringe cutting seam carving algorithm to find fringe lines to obtain a fringe seam map, and then loads the fringe seam map into the reference image database;

The image size adjustment unit inputs the fringe seam image from the fringe search unit, and successively removes a fringe with the lowest energy according to the target size preloaded into the computer to obtain the rear image;

The corresponding point calculation unit loads the stripe seam map from the stripe search unit, and calculates the positions of the points on both sides of the stripes in the original image in the corresponding points in the rear image: initially, each point on the original image The position of the corresponding point is the current position, and each time a stripe is removed: if the removed stripe is a vertical stripe, the corresponding point position of the point to the left of the stripe remains unchanged; if the removed stripe is a horizontal stripe, Then the position of the corresponding point of the point above the stripe remains unchanged; if the removed stripe is a vertical stripe, the position of the corresponding point of the point on the right side of the stripe moves to the left by one unit in the horizontal direction; if removed If the stripe is a horizontal stripe, the position of the corresponding point of the point below the stripe moves upward by one unit in the vertical direction;

The fidelity evaluation unit compares the difference between the point on the original image and the window matrix of the corresponding point on the subsequent image, so as to calculate the fidelity value of the latter image, and obtain the fidelity evaluation based on the quantitative result result;

The window matrix of a certain point refers to: a 9*9 rectangular window centered on a certain point on the graph where the point is located, and its content is the signal value at the midpoint of the window;

Step (2): The computer carries out the described fidelity evaluation method according to the following steps successively,

Step (2.1), from the energy calculation unit according to the energy function of the input needs evaluation, the original image output energy map; from the stripe search unit using the stripe cutting algorithm to output the stripe seam graph according to the energy graph; from the adjustment image size unit according to the setting The target size and the striped seam figure output the back figure to the fidelity evaluation unit, and each point between each point on the original picture and the position of the point in the back figure is output from the corresponding point calculation unit Corresponding to the fidelity evaluation unit;

Step (2.2), the fidelity evaluation unit performs the fidelity evaluation according to the following steps:

Step (2.2.1), converting the loaded back image from the red, green and blue primary color space RGB to the lαβ color space, wherein l represents the brightness channel of non-color, α represents the yellow and blue channel of color, and β represents the red color of color Green channel, find all the points on both sides of the stripe; if one of the four points above, below, left, and right of a certain point belongs to a certain stripe, then the certain point is the point on both sides of the certain stripe;

Step (2.2.2), calculating the fidelity value Q of a single color channel of the 9*9 matrix window of the single corresponding point of the points on both sides of the stripe obtained in step (2.2.1):

$\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; xy</span>}}}{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; xy</span>}}}{{<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}\frac{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}{{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

Among them, x is the signal value of a point in the window matrix of the original image on a certain channel; y is the signal value of a point in the window matrix of the corresponding point in the following figure on this channel, and x, y are in the window matrix The average value of the signal value of the point on a certain channel, δx, δy is the variance, δxy is the covariance;

Step (2.2.3), the fidelity value of the corresponding point on all described back figure to be calculated is carried out arithmetic mean respectively by 1, α, β three channels, obtains $Q_z=\left(\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{Q}_i\right)/m$ Wherein, z=l, α, β, M is the number of corresponding points;

In step (2.2.4), the fidelity values Q _l , Q α , and Q β of the corresponding points of the three channels l, _α , and _β are weighted and geometrically averaged to obtain the fidelity value of the entire image $Q_{\mathrm{color}}=\sqrt{{\mathrm{\&omega;}}_l{Q}_l^2+{\mathrm{\&omega;}}_{\mathrm{\&alpha;}}{Q_{\mathrm{\&alpha;}}}^2+{\mathrm{\&omega;}}_{\mathrm{\&beta;}}{Q_{\mathrm{\&beta;}}}^2}.$

It can be seen from the above method that using the energy function that needs to be evaluated, the content-related size scaling of the pre-selected image is performed through the seam carving algorithm, and the corresponding relationship between the positions of the original image and the post-image is calculated during the process of adjusting the image , and finally compare the point on the original image with the window matrix of the corresponding point to obtain the quantified image fidelity evaluation result, which reduces the evaluation delay and obtains an objective and accurate evaluation effect.

Description of drawings

Figure 1 Exemplary structure diagram of the fidelity evaluation system

Figure 2 Schematic diagram of the evaluation process of the fidelity evaluation system

Fig. 3 Exemplary flowchart of fidelity evaluation method

The flow chart of the fidelity method in the embodiment of Fig. 4

Figure 5 Schematic diagram of striped seam line and corresponding point matrix window

Detailed ways

According to the above-mentioned first main purpose, the present invention provides a method for evaluating image fidelity, comprising the following steps:

A uses the energy function that needs to be evaluated to perform content-dependent size scaling on pre-selected images through the seam carving algorithm.

B finds the corresponding points in the original image in the latter image.

C compares the latter image with the original image, and obtains the fidelity evaluation result of the energy function according to the comparison result.

The content-related scaling of the image in step A is as follows: After selecting the subject content to be retained on the image, using the energy function to be evaluated, the image is scaled to any size based on the subject content through the seam carving algorithm.

Find the corresponding relationship between the positions of the points on the original image and the subsequent image as described in step B: while adjusting the size of the original image by increasing or decreasing the seam line, find the corresponding points of each point on the original image on the subsequent image. .

In step C, the evaluation result obtained according to the comparison result is: only compare the difference between the points on the adjacent sides of the seam line in the original image and the window matrix of the corresponding point on the subsequent image, so as to calculate the matching value of the latter image and obtain a quantitative comparison result.

According to the above second main purpose, the present invention provides an image fidelity evaluation system, comprising: a reference image database, a seam carving algorithm unit, and a fidelity evaluation unit.

The reference image database provides reference image material to be processed. These include: original image, energy image, and seam image. The reason for this design is to protect the author's intellectual property rights of certain energy functions. That is to say, if the original author is unwilling to provide the code of the energy function, he can only provide the energy map or the seam pattern of the original image, and the evaluation can also be realized.

The seam carving algorithm unit will load the energy function to be measured, and perform content-related image size adjustment on the reference image through the seam carving algorithm. At the same time, the corresponding points of each point in the original image on the subsequent image will be recorded during the adjustment process.

The fidelity evaluation unit calculates the degree of matching between the rear image and the original image, and compares the difference between the points on both sides of the seam line of the original image and the window matrix of the corresponding point by using the image change characteristics, thereby calculating the retention of the rear image. The truth value, get the quantified comparison result and get the evaluation result of the algorithm fidelity according to the comparison result.

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and examples.

The basic idea of the fidelity evaluation method in the present invention is to use the energy function to be evaluated to perform content-related size scaling on the pre-selected image through the seam carving algorithm. Find the corresponding points in the original image in the latter image. Compare the latter image with the original image, and obtain the fidelity evaluation result of the energy function according to the comparison result. As shown in Figure 3

FIG. 1 is an exemplary structural diagram of an image fidelity evaluation system in the present invention. As shown in FIG. 1 , the image guarantee evaluation system of the present invention includes: a reference image database 11 , a seam carving algorithm unit 12 , and a fidelity evaluation unit 13 . The seamcarving algorithm unit itself includes an energy calculation unit 121 , a seam search unit 122 , a size adjustment unit 123 , and a corresponding point calculation unit 124 .

Fig. 2 is a schematic diagram of the evaluation process of the fidelity evaluation system in the present invention. As shown in Figure 2, the evaluation process of the present invention is:

The reference image database 11 provides pre-stored reference images to the seam carving algorithm unit 12 and the fidelity evaluation unit 13, and the reference image database may include ordinary images, energy images and fringe seam images.

The seam carving algorithm unit 12 uses the energy function to be evaluated to perform content-related size scaling on the pre-selected image through the seam carving algorithm, and at the same time finds the corresponding point in the original image in the subsequent image. Provide the corresponding relationship between each point of the rear image and the original image to the fidelity evaluation unit 13; wherein the energy calculation unit 121 uses the energy function to calculate from the color information of the image to obtain the energy map of the original image, and skip 121 if the energy map has been provided unit; looking for the seam unit 122 using the seam carving algorithm to calculate from the energy map to obtain the striped seam map, if the striped seam map has been provided, the two units 121 and 122 are skipped. The resizing unit 123 adjusts the size of the image by increasing or decreasing the seam line according to the fringe seam image and the target size to obtain the rear image. The corresponding point calculation unit 124 calculates the corresponding relationship between the rear image and the original image, and obtains the corresponding points of each point in the original image on the latter image.

The fidelity evaluation unit 13 will use the corresponding relationship between the original image, the rear image and the points, compare the window matrix of the corresponding points on the rear image and the original image, and obtain the fidelity of the seam carving algorithm under the energy function to be measured according to the comparison result. Degree assessment results.

In the following, the image fidelity evaluation method in the present invention will be described in detail in combination with specific embodiments.

Step 41, initialization.

Load the original image, select the energy function, and set the target size.

Step 42, energy calculation.

Use the energy function to be evaluated to calculate the energy of each point in the original image, and generate an energy graph. Step 43 is to search for seam.

According to the energy map provided in the previous step, use the seam carving algorithm to find the seam line, and then get the striped seam map

Step 44, adjust image size.

According to the target size, the seam line with the lowest current energy is removed each time. After getting the picture

Step 45, corresponding point calculation.

Initially, the corresponding point position of each point is the current position of the point. Every time a seam is removed, the corresponding point position of the point on the left side of the seam (vertical seam) or the top side (horizontal seam) remains unchanged, and the right side of the seam (vertical seam) or bottom side (Horizontal seam) The corresponding point position of the point is moved left (vertical seam) or up (horizontal seam) by one unit. This step is performed alternately with step 44.

Step 46, fidelity evaluation.

First convert the image from the RGB color space to the lαβ color space (where l represents the brightness channel of non-color, α represents the yellow and blue channel of color, and β represents the red and green channel of color), and then find all the points on both sides of the seam . Since seam has horizontal and vertical, the definition of points on both sides of seam is: if one of the four points above, below, left, and right of a certain point belongs to a certain seam, then this point is counted as the point on both sides of seam. Calculate the fidelity value of a single color channel of a single corresponding point 9×9 rectangular window (as shown in Figure 5) of the points on both sides of the seam. The calculation method of this fidelity value is as follows:

$\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; xy</span>}}}{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; xy</span>}}}{{<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}\frac{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}{{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

Where x is the signal value of a point in the window matrix of the original image on a certain channel, y is the signal value of a point in the window matrix of the corresponding point in the subsequent image on this channel, x, y are the points in the window matrix in a certain channel The mean value of the signal value on a channel, δx, δy is the variance, δxy is the covariance.

Then average the fidelity values of each channel for all corresponding points to be calculated to obtain

$Q_z=\left(\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{Q}_i\right)/m,(z=l,\mathrm{\&alpha;},\mathrm{\&beta;}),$ The standard value is 1 per channel.

Then the weighted average of the corresponding point fidelity values of the three channels is obtained

${\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; color</span>}}<span\; class="notranslate">\; =</span>\sqrt{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}{{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}{{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

This is the fidelity value of the whole image. Now the coefficient of each channel is 1, so the standard value of the total Q value is

比较，越接近标准值，说明被评测算法的保真度效果越好。Compare the resulting fidelity value with the standard value

Comparison, the closer to the standard value, the better the fidelity effect of the evaluated algorithm.

Wherein, the calculation formula is adopted in step 46 $Q=\frac{{\mathrm{\&delta;}}_{\mathrm{xy}}}{{\mathrm{\&delta;}}_x{\mathrm{\&delta;}}_{\mathrm{the\; y}}}\frac{2\stackrel{\&OverBar;}{\mathrm{xy}}}{{\left(\stackrel{\&OverBar;}{x}\right)}^2+{\left(\stackrel{\&OverBar;}{\mathrm{the\; y}}\right)}^2}\frac{2{\mathrm{\&delta;}}_x{\mathrm{\&delta;}}_{\mathrm{the\; y}}}{{{\mathrm{\&delta;}}_x}^2+{{\mathrm{\&delta;}}_{\mathrm{the\; y}}}^2},$ It is because it is simple to calculate and has a strong correlation with subjective quality evaluation, and can obtain a total similarity measure from the three sub-aspects of brightness, contrast, and structure as an objective quality evaluation standard. This method fully considers the structural information of the image and the characteristics of human vision. Starting from the understanding function of the image content, the subjective visual perception of the human eye on the image quality is estimated through mathematical modeling, so that the structural similarity calculation model meets the requirements of image processing applications. Nature. For this formula, see [Zhou Wang and Alan C.Bovik, AUniversal Image Quality Index, IEEE Signal Processing Letters, vol.9, no.3, pp.81-84, March2002]

Claims (2)

1. A method for evaluating fidelity based on content-related image size adjustment, characterized in that the method is implemented in a computer according to the following steps: Step (1): Initialize, set up the following units in the computer to build an evaluation system: The reference image database unit stores reference image data to be processed, including: original image, energy image, and fringe seam image; The strip cutting algorithm unit, after inputting the energy function to be measured, performs content-related image size adjustment on the loaded reference image through the strip cutting seam carving algorithm, and records the position of each point in the original image during the adjustment process. The position of the corresponding point on the rear map obtained after the adjustment is completed, wherein there are: an energy calculation unit, a fringe search unit, an image size adjustment unit and a corresponding point calculation unit, wherein: The energy calculation unit loads the energy function to be evaluated and loads the original image from the reference image database, calculates the energy of points in the original image to generate an energy map, and then loads the energy map into the reference image database; The fringe search unit loads the energy map from the energy calculation unit, uses the fringe cutting seam carving algorithm to find fringe lines to obtain a fringe seam map, and then loads the fringe seam map into the reference image database; The image size adjustment unit inputs the fringe seam image from the fringe search unit, and successively removes a fringe with the lowest energy according to the target size preloaded into the computer to obtain the rear image; The corresponding point calculation unit loads the stripe seam map from the stripe search unit, and calculates the positions of the points on both sides of the stripes in the original image in the corresponding points in the rear image: initially, each point on the original image The position of the corresponding point is the current position, and each time a stripe is removed: if the removed stripe is a vertical stripe, the corresponding point position of the point to the left of the stripe remains unchanged; if the removed stripe is a horizontal stripe, Then the position of the corresponding point of the point above the stripe remains unchanged; if the removed stripe is a vertical stripe, the position of the corresponding point of the point on the right side of the stripe moves to the left by one unit in the horizontal direction; if removed If the stripe is a horizontal stripe, the position of the corresponding point of the point below the stripe moves upward by one unit in the vertical direction; The fidelity evaluation unit compares the difference between the point on the original image and the window matrix of the corresponding point on the subsequent image, so as to calculate the fidelity value of the latter image, and obtain the fidelity evaluation based on the quantitative result result; The window matrix of a certain point refers to: a 9*9 rectangular window centered on a certain point on the graph where the point is located, and its content is the signal value at the midpoint of the window; Step (2): The computer carries out the described fidelity evaluation method according to the following steps successively, Step (2.1), from the energy calculation unit according to the energy function of the input needs evaluation, the original image output energy map; from the stripe search unit using the stripe cutting algorithm to output the stripe seam graph according to the energy graph; from the adjustment image size unit according to the setting The target size and the striped seam figure output the back figure to the fidelity evaluation unit, and each point between each point on the original picture and the position of the point in the back figure is output from the corresponding point calculation unit Corresponding to the fidelity evaluation unit; Step (2.2), the fidelity evaluation unit performs the fidelity evaluation according to the following steps: Step (2.2.1), convert the loaded image from the red, green and blue primary color space RGB to the |αβ color space, where | represents the brightness channel of non-color, α represents the yellow and blue channel of color, and β represents the color space Red and green channels, find all the points on both sides of the stripe; if one of the four points above, below, left, and right of a certain point belongs to a certain stripe, then the certain point is the point on both sides of the certain stripe; Step (2.2.2), calculating the fidelity value Q of a single color channel of the 9*9 matrix window of the single corresponding point of the points on both sides of the stripe obtained in step (2.2.1): $\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; xy</span>}}}{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; xy</span>}}}{{<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}\frac{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}{{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$ Among them, x is the signal value of a point in the window matrix of the original image on a certain channel; y is the signal value of a point in the window matrix of the corresponding point in the following figure on this channel, and x, y are in the window matrix The average value of the signal value of the point on a certain channel, δx, δy is the variance, δxy is the covariance; Step (2.2.3), the fidelity value of the corresponding point on all the described back graphs to be calculated is carried out arithmetic mean respectively according to |, α, β three channels, obtains $Q_z=\left(\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{Q}_i\right)/m$ Wherein, z=l, α, β, M is the number of corresponding points; In step (2.2.4), the fidelity values Q _l , Q α , and Q β of the corresponding points of the three channels |, _α , and _β are weighted and geometrically averaged to obtain the fidelity value of the entire image $Q_{\mathrm{color}}=\sqrt{{\mathrm{\&omega;}}_l{Q_l}^2+{\mathrm{\&omega;}}_{\mathrm{\&alpha;}}{Q_{\mathrm{\&alpha;}}}^2+{\mathrm{\&omega;}}_{\mathrm{\&beta;}}{Q_{\mathrm{\&beta;}}}^2}.$ 2. The fidelity evaluation method based on content-related image resizing, which is characterized by the difference from 1 in that in step (2.2.4), the fidelity values Q _l , Q _α , Q _β directly according to $Q_{\mathrm{color}}=\sqrt{{Q_l}^2+{Q_{\mathrm{\&alpha;}}}^2+{Q_{\mathrm{\&beta;}}}^2}$ Get the fidelity value of the whole image, other features and steps are the same as 1.

Images