CN103873854B - The defining method of a kind of stereo-picture subjective assessment subject's quantity and experimental data - Google Patents

Abstract

The invention discloses a method for determining the number of subjects for subjective evaluation of stereoscopic images and experimental data, and relates to the field of subjective evaluation of stereoscopic image quality. Quantity, so as to shorten the experimental cycle and save costs; use the above-mentioned reasonable number of qualified subjects to conduct subjective evaluation experiments, including establishing a stereoscopic image database, selecting reasonable test environment conditions, selecting subjects according to standards, and implementing test subjects. In the process of training and testing, in the end, according to the subjective evaluation standard of image quality and the Grubbs test method, abnormal subjects are eliminated and abnormal data in the experimental data are reasonably screened to avoid uncertain factors such as subjects and experimental environment influence on the accuracy of experimental results. Based on the subjective evaluation of stereoscopic image quality, this method improves the accuracy of experimental data results from the perspective of probability and statistics.

Description

A method for determining the number of subjects and experimental data for subjective evaluation of stereoscopic images

technical field

The invention relates to the field of stereoscopic images, in particular to a method for determining the number of subjects and experimental data for subjective evaluation of stereoscopic images.

Background technique

At present, stereoscopic imaging technology has been widely used in various industries. This technology includes the acquisition, compression, transmission, processing and display of stereoscopic images. Therefore, how to use stereoscopic image quality evaluation methods to measure the degree of image damage caused by stereoscopic imaging in these processes has become one of the research hotspots in the field of stereoscopic imaging technology. Stereoscopic image quality evaluation is divided into two categories: subjective evaluation and objective evaluation. Compared with objective evaluation methods, subjective evaluation methods directly use people to judge the quality of stereoscopic images, which can more intuitively and accurately reflect the true quality of stereoscopic images. Subjective quality evaluation must go through many links, mainly including the establishment of a stereoscopic image database, the selection of test environment conditions, the selection of subjects, the training and test process, the screening and processing of experimental data, etc. However, factors such as the uncertainty of the experimental environment and the status of the subjects will directly affect the accuracy and precision of the subjective evaluation of the experimental data. At the same time, the number of subjects will directly affect the experimental testing time and subjective experimental costs. Therefore, it is very important to propose a feasible method for determining the number of subjects and screening methods for experimental data.

In recent years, domestic and foreign scholars have begun to study the subjective evaluation scheme of stereoscopic image quality. Based on the subjective evaluation standard ITU-R BT.500-11 of the planar image quality, the International Telecommunication Union ITU proposed the subjective evaluation standard ITU-R BT.1438 ^[1] for stereoscopic TV images. A simple and general introduction to the test process and the test process; literature [2], [3], [4] define the subjective evaluation scheme from the aspects of viewing conditions, viewing personnel, testing methods, etc.; the subjective evaluation method proposed in the literature [5] establishes The literature [6] analyzed various factors that affect the quality of stereo perception from the subjective aspect, and discussed the characteristics of the existing stereo image databases; literature [7] from Establishing a stereoscopic image database, selection of subjects, experimental conditions, training and testing of subjects, etc., established a relatively complete subjective evaluation scheme of stereoscopic image quality. Different from the stereoscopic image quality subjective evaluation scheme, literature [8] realizes a human-computer interaction stereoscopic video quality subjective testing system through software programming; the stereoscopic video quality subjective evaluation method proposed in literature [9] adopts SPSS tool for data process and compare. In addition, the subjective evaluation method of stereoscopic image quality is often used to study various factors and characteristics related to stereoscopic imaging. Literature [10] explores the influence of right view quality and disparity image quality on depth perception through subjective experiments; literature [11] uses subjective The experiment explores the relationship between brightness and stereo perception.

The inventor finds in the process of realizing the present invention that the main shortcoming of the prior art is as follows:

At present, there is no uniform standard for the subjective evaluation scheme of stereoscopic images. At the same time, in the established or used subjective evaluation schemes of stereoscopic image quality, there are no specific regulations on the method of determining the number of subjects and the screening method of experimental data. The number of subjects and the accuracy of experimental data directly determine the complexity, cost, and accuracy of experimental results of subjective experiments, and are also important factors affecting subjective evaluation scores.

Contents of the invention

The present invention provides a method for determining the number of subjects and experimental data for subjective evaluation of three-dimensional images. This method improves the correctness of the subjective evaluation results. See the following description for details:

A method for determining the number of subjects and experimental data for subjective evaluation of stereo images, said method comprising the following steps:

(1) Select n subjects to pre-evaluate the stereoscopic image quality, obtain subjective evaluation scores, and calculate the average value and standard deviation;

(2) According to the required accuracy and confidence requirements, the number of subjects N actually required for the subjective experiment is obtained through iterative operations by inverting the confidence interval formula;

(3) Select N subjects for subjective evaluation of stereoscopic image quality, and calculate the average and standard deviation of the evaluation scores for each test image;

(4) According to whether the evaluation score of each test image satisfies the normal distribution, different thresholds p _i and q _i for judging abnormal scores are used to calculate the index for judging abnormal data and Eliminate abnormal subjects by comparison;

(5) Use the Grubbs test method to screen out abnormal data in the evaluation scores of each test image;

(6) Using the screened subjects and data without abnormalities, calculate the average value of all normal experimental data of each test image to obtain the final subjective evaluation result.

The determination of the number of subjects is as follows:

N＝(t ² S ² )/(r ² mean ² ) (1)

Among them, r represents the acceptable relative percentage deviation, mean and S represent the average value and standard deviation of the subjective evaluation scores of a small number of subjects who pre-evaluate the quality of a single stereoscopic image, and the initial value of t is 1.96, which is substituted into the formula ( 1) Calculate the initial value of the number of subjects N, and then substitute the t value corresponding to the initial value of N (obtained from the t distribution table) into formula (1) to calculate the N value, and iterate continuously until the calculated N value does not exceed Another change occurs, namely the number of subjects.

The abnormal data in the evaluation scores of each test image screened out by using the Grubbs test method is specifically:

First, the evaluation scores of each test image are arranged in ascending order; parameters α _i and γ _i are calculated according to formulas (2) and (3),

α _i =(x _imax -u _i )/S _i (2)

γ _i ＝(u _i -x _imin )/S _i (3)

Among them, x _imax and x _imin represent the maximum value and minimum value of all evaluation scores of the i-th test image respectively; u _i and S _i represent the average value and standard deviation of the i-th test image evaluation scores respectively; finally, pass Check the critical value table to obtain the critical value G corresponding to the number of subjects and a confidence level of 95%. If α _i >G, then eliminate the abnormal data x _imax , and if γ _i >G, then eliminate the abnormal data x _imin ; finally, repeat The above inspection steps are performed until there are no abnormal data in the experimental data.

The beneficial effects of the technical solution provided by the invention are: the method perfects the subjective evaluation scheme of the stereoscopic image quality, and compared with the prior art, the method focuses on the method of determining the number of subjects and screening experimental data. The experimental results show that this method is feasible, which not only ensures the reliability of the experimental results, but also saves experimental resources and improves the correctness of the subjective evaluation results.

Description of drawings

Figure 1 is the left and right viewpoints of the girl reference image;

Figure 2 is the left and right viewpoints of the family reference image;

Figure 3 is the left and right viewpoints of the boy reference image;

Figure 4 is the left and right viewpoints of the flower reference image;

Figure 5 is the left and right viewpoints of the tree reference image;

Figure 6 is the left and right viewpoints of the river reference image;

FIG. 7 is a flow chart of a subjective evaluation scheme for stereoscopic image quality;

Figure 8 is a 3D WINDOWS-19A0 computer stereoscopic imaging device.

Detailed ways

In order to make the purpose, technical solution and advantages of the present invention clearer, the implementation manners of the present invention will be further described in detail below.

In order to improve the correctness of the subjective evaluation results, the embodiment of the present invention provides a method for determining the number of subjects and experimental data for the subjective evaluation of the stereoscopic image. The left and right viewpoints of the reference stereoscopic image are shown in Figures 1-6. See Figure 7 for the evaluation plan process, and see the description below for details:

101: Select n subjects to pre-evaluate the stereoscopic image quality, obtain subjective evaluation scores, and calculate the average value and standard deviation;

First, use a relatively complete stereoscopic image quality subjective evaluation system to conduct a pre-test on a small number of selected subjects, including establishing a stereoscopic image database, determining viewing conditions, selecting subjects according to standards, and conducting training and testing ^[7] , and finally get the subjective evaluation scores {x ₁ ,x ₂ ,...x _n } of n subjects who pre-evaluate the quality of a single stereo image, and calculate their mean values according to formula (1) and formula (2) and standard deviation,

$\mathrm{<span\; class="notranslate">\; mean</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; =</span>\sqrt{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; mean</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

102: According to the required accuracy and confidence requirements, the number of subjects N actually required for the subjective experiment is obtained through iterative operations by inverting the confidence interval formula;

Calculate the number of subjects N according to formula (3),

N＝(t ² S ² )/(r ² mean ² ) (3)

Among them, r represents the acceptable relative percentage deviation, mean and S represent the average value and standard deviation of the subjective evaluation scores of a small number of subjects who pre-evaluate the quality of a single stereoscopic image; t first takes the initial value of 1.96 and substitutes it into the formula (3) Calculate the initial value of the number of subjects N, and then use the t value corresponding to the initial value of N (obtained from the t distribution table) to be substituted into the formula (3) to calculate the N value, iteratively, and finally, until the calculation The N value out will no longer change, that is, the number of subjects.

103: Select N subjects to perform subjective evaluation of stereoscopic image quality, and calculate the average value and standard deviation of the evaluation scores of each test image;

First, assuming that a total of I test images are used, N subjects are set to participate in the subjective experiment according to the above calculation, and the test is carried out again through a relatively complete subjective evaluation system of stereoscopic image quality, x _i,j represents the jth subject For the evaluation score of the i-th test image, calculate the average value and standard deviation of each test image evaluation score, such as formula (4) and formula (5),

${\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

Among them, u _i and S _i represent the average value and standard deviation of the i-th test image evaluation scores, respectively.

104: According to whether the evaluation score of each test image satisfies the normal distribution, different thresholds p _i and q _i for judging abnormal scores are used to calculate the index for judging abnormal data and Eliminate abnormal subjects by comparison;

First, calculate the second-order moment and fourth-order moment of the evaluation score of the i-th test image according to formulas (6) and (7) respectively, and obtain the kurtosis coefficient of the evaluation score of the i-th test image, such as formula (8), according to According to the requirements of ITU-R BT.500-11, if 2≤β _2i ≤4, the evaluation result of the i-th test image is a normal distribution; otherwise, it is a non-normal distribution.

${\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

For each test image, calculate the thresholds p _i and q _i for the subjects to judge whether the scores are abnormal. If the evaluation scores conform to the normal distribution, use formulas (9) and (10) to calculate; otherwise, use formulas (11) and (12) calculation,

p _i =u _i +2S _i (9)

q _i =u _i -2S _i (10)

${\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>\sqrt{\mathrm{<span\; class="notranslate">\; 20</span>}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\sqrt{\mathrm{<span\; class="notranslate">\; 20</span>}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 12</span>}<span\; class="notranslate">\; )</span>$

Then, classify the judgment scores of all test images by the jth subject, and obtain the measurement parameters P _j and Q _j for judging two types of situations. That is, the initial values are all set to 0, and all evaluation scores x _i,j (i=1,2,...,I) of the jth subject are compared with p _i and q _i of the corresponding test image; If the evaluation score of the i-th test image conforms to the normal distribution, judge according to formulas (9) and (10), if x _i,j ≥ p _i , then P _j =P _j +1, if x _i,j ≤ q _i , then Q _j =Q _j +1; otherwise, judge according to formulas (11) and (12).

Finally, according to formula (13) and formula (14), the index for judging abnormal data is calculated and like and Then eliminate all the evaluation data of the jth subject, otherwise, keep the evaluation data.

${\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; I</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 13</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; |</span>\frac{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 14</span>}<span\; class="notranslate">\; )</span>$

105: Use the Grubbs test method to filter out the abnormal data in the evaluation score of each test image;

First, the evaluation scores of each test image are arranged in ascending order, and the maximum and minimum values of all evaluation scores of the i-th test image are x _imax and x _imin , according to formulas (15) and (16) Calculate the parameters α _i and γ _i , where u _i and S _i represent the mean value and standard deviation of the evaluation scores of the i-th test image, respectively.

α _i =(x _imax -u _i )/S _i (15)

γ _i ＝(u _i -x _imin )/S _i (16)

Finally, the critical value G corresponding to the number of subjects and a confidence level of 95% is obtained by checking the critical value table. If α _i >G, the abnormal data x _imax will be eliminated, and if γ _i >G, the abnormal data x _imin will be eliminated. Afterwards, the above inspection steps are repeated until there is no abnormal data in the experimental data.

106: Using the screened subjects and data without abnormalities, calculate the average value (ie MOS value) of all normal experimental data of each test image to obtain the final subjective evaluation result.

According to the formula (17), calculate the average value (ie, MOS value) of all normal experimental data of each test image to obtain the final subjective evaluation result.

${\mathrm{<span\; class="notranslate">\; MOS</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 17</span>}<span\; class="notranslate">\; )</span>$

Among them, MOS _i represents the subjective evaluation value of the i-th image, K represents the number of final normal subjects, and _xi,j represents the evaluation score of the j-th subject on the i-th image.

In order to verify the algorithm for determining the number of subjects and the experimental data screening algorithm proposed by this method, this method subjectively evaluates images distorted by JPEG compression, random noise, and Gaussian noise. The following is a brief introduction to the experimental process:

This method selects 48 test images for the pre-test to determine the number of subjects, including six scenes of girl, family, boy, flower, tree and river, which are respectively tested by 85%, 65%, 45%, and 25%. , 15%, 5% JPEG compression and adding random noise and Gaussian noise; the left and right viewpoint images of six reference stereo images are shown in Figures 1 to 6, and the experimental data are provided by the Institute of Broadband Wireless Communication and Stereo Imaging of Tianjin University. According to the ITU-R BT.1438[1] standard, the experiment adopts DSIS (Double-Stimulus Impairment Scale, double-stimulus damage scale method), recruiting 15 subjects with normal visual function and stereo vision to participate in the pre-test, watching The equipment is 3DWINDOWS-19A0 (19 inches, 1280×1024), as shown in Figure 8, the brightness of the laboratory is extremely weak, and the quality of the distorted image is scored according to the DSIS scoring standard. The total time of training and testing for each subject cannot more than 30 minutes.

Result verification:

In this embodiment, the algorithm for determining the number of subjects introduced in this patent will be used to calculate the number of subjects required for different test images (see Table 1 and Table 2 for specific results), where the confidence level is set to 95%, due to poor quality The evaluation scores of images fluctuate greatly, so the acceptable percentage deviation r is set to 5% for pre-test images with an average evaluation score greater than 3, and 8% for other images.

Table 1 and Table 2 respectively show some experimental data of JPEG compressed and noise-added test scenes. It can be seen from observation that the number of subjects is related to scene content, image processing method and image damage degree. Through the calculation of the algorithm for determining the number of subjects, for the stereoscopic image database established in this patent, 30 subjects were selected to participate in the subjective evaluation experiment, which not only meets the requirements of most test images for the number of subjects, but also ensures the experimental results The accuracy and reliability of the experiment will not be wasted due to the large number of subjects. In order to further verify the correctness of this method and that 30 subjects can ensure the accuracy of the experimental results, this method increases the number of subjects to 35 to conduct subjective evaluation experiments on some test images, and found that the experimental results are consistent with those of 30 subjects. The measured results are very close, see Table 1 and Table 2 for details.

Table 1 Algorithm verification results on compressed images

Table 2 Algorithm verification results on noised images

Abnormal and suspicious subjects and experimental data are eliminated by this method, so that the experimental results are more accurate and reliable. In order to verify the correctness and effectiveness of the experimental data screening method, this embodiment compares the MOS value of the final evaluation result of the subjective experiment with the partial evaluation scores of the eliminated subjects, see Table 3 for details. It can be seen from observation that the difference between the evaluation scores of the eliminated subjects and the final MOS value of the subjective experiment is more than 1.5 points, which deviates from the evaluation score range of most of the subjects and will affect the accuracy of the subjective evaluation results, thus verifying the patent The correctness of the experimental data screening method. Therefore, this method can correctly determine the number of subjects and screen experimental data, ensure the reliability of experimental results, save experimental resources, and eliminate abnormal and suspicious experimental values, thereby ensuring the consistency of experimental results.

Table 3 Verification results about the data screening algorithm

references:

[1]International Telecommunication Union(ITU),Recommendation ITU-R BT.1438,Subjective assessment of stereoscopic television pictures[S],2000.

[2]W.J.Tam, G.Alain, L.Zhang, et al.Smoothing depth maps for improved stereoscopic image quality[C].Three-Dimensional TV,Video and DisplayⅢ,Philadelphia,PA,USA,2004.

[3] Xu Wang, Mei Yu, You Yang, et al. Research on subjective stereoscopic image quality assessment [C]. Multimedia Content Access: Algorithms and SystemsⅢ, San Jose, CA, United States, 2009.

[4]CHENG Yu-qing and JIANG Xiu-hua.Latest research result of subjective and objective evaluation methodology for the video quality of 3DTV[J].Journal of communicationuniversity of China(science and technology).2012,19(1):31- 36.

[5] Junming Zhou, Gangyi Jiang and Xiangying Mao. Subjective quality analyzes of stereoscopic images in 3DTV system [C], IEEE Visual Communications and Image Processing (VCIP), Tainan: 2011, 1～4

[6] Zhou Wujie, Yu Mei, Zhou Junming, etc., Research on Stereoscopic Image Quality Evaluation Method [C], 2010 China International Conference on Communications, Nanning: 2010

[7] Zhang Yingjing, Li Sumei, Wei Jinjin, etc., Subjective Evaluation Scheme of Stereoscopic Image Quality [J], Acta Photonica Sinica, 2012, 41(5): 602～607

[8] NukhetOzbek, GizemErtan and OktayKarakus. Interactive quality assessment for asymmetric coding of 3D video[C], 3DTV Conference: The True Vision-Capture, Transmission and Display of 3D Video (3DTV-CVON), Antalya: 2011, 1～4

[9] A.S.Umar, R.M.Swash and A.H.Sadka, Subjective quality assessment of 3D videos[C], 2011AFRICON, Livingstone, Zambia: 2011: 1～6

[10] ChaohuiLv, Jingwei Huang and YinghuaShen, Subjective assessment of noised stereoimages[C], International Conference on Multimedia Technology (ICMT), Hangzhou, China: 2011, 783～785

[11] Mahsa T.Pourazad, Zicong Mai, Panos Nasiopoulos, et al, Effect of brightness on the quality of visual3D perception[C], 201118th IEEE International Conference on Image Processing (ICIP), Brussels: 2011, 989～992

Those skilled in the art can understand that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the serial numbers of the above-mentioned embodiments of the present invention are for description only, and do not represent the advantages and disadvantages of the embodiments.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (1)

1. a defining method for stereo-picture subjective assessment subject quantity and experimental data, is characterized in that, said method comprising the steps of:

(1) choose n name subject and carry out stereo image quality pre assessment, obtain subjective assessment mark, calculating mean value and standard deviation;

(2) according to the requirement of required accuracy and confidence level, by inverse confidential interval formula, interative computation obtains the subject quantity N of subjective experiment actual needs;

(3) choose N name subject and carry out stereo image quality subjective assessment, calculate mean value and the standard deviation of every width test pattern evaluation score;

(4) whether meet normal distribution according to the mean value of every width test pattern evaluation score and standard deviation, adopt the threshold value p of different judge mark exceptions _iand q _i, calculate the index passing judgment on abnormal data with by comparing, eliminate abnormal subject;

(5) employing Grubbs' test method filters out the abnormal data in every width test pattern evaluation score;

(6) there are not abnormal subject and data after utilizing screening, the mean value calculating whole normally experimental datas of every width test pattern obtains final subjective evaluation result;

Wherein, described subject's quantity is determined to be specially:

N＝(t ²S ²)/(r ²mean ²) (1)

Wherein, r represents acceptable percentage deviation, mean, S represent that a small amount of subject carries out mean value and the standard deviation of the subjective assessment mark of pre assessment to single stereoscopic image quality respectively, t gets initial value 1.96, substitute into the initial value that formula (1) calculates subject quantity N, then substitute into formula (1) calculating N value by the t value corresponding to N initial value, continuous iteration, until the N value calculated no longer changes, i.e. subject's quantity;

Wherein, the abnormal data that described employing Grubbs' test method filters out in every width test pattern evaluation score is specially:

First, the evaluation score of every width test pattern is arranged in ascending order mode; According to formula (2) and (3) calculating parameter α _iand γ _i,

α _i＝(x _imax-u _i)/S _i(2)

γ _i＝(u _i-x _imin)/S _i(3)

Wherein, x _imaxand x _iminrepresent the maximum in whole evaluation score of the i-th width test pattern and minimum value respectively; u _iand S _irepresent mean value and the standard deviation of the i-th width test pattern evaluation score respectively; Finally, by looking into tables of critical values and obtain corresponding subject's quantity and confidence level being the critical value G of 95%, if α _i> G, then eliminate abnormal data x _imaxif, γ _i> G, then eliminate abnormal data x _imin; Finally, above-mentioned checking procedure is repeated until no longer there is abnormal data in experimental data;

Wherein, the threshold value p of different judge mark exceptions is adopted _iand q _i, calculate the index passing judgment on abnormal data with by comparing, the step of eliminating abnormal subject is specially:

For every width test pattern, calculate subject and pass judgment on the whether abnormal threshold value p of mark _iand q _iif evaluation score meets normal distribution, then formula (4) and (5) are utilized to calculate; Otherwise, utilize formula (6) and (7) to calculate,

p _i＝u _i+2S _i(4)

q _i＝u _i-2S _i(5)

$p_i=u_i+\sqrt{20}{S}_i---\left(6\right)$

$q_i=u_i-\sqrt{20}{S}_i---\left(7\right)$

The index passing judgment on abnormal data is calculated according to formula (8) and formula (9) with if and then eliminate all evaluating datas of jth name subject, otherwise, retain evaluating data;

$R_j^1=\frac{1}{I}(P_j+Q_j)---\left(8\right)$

$R_j^2=\left|\frac{P_j-Q_j}{P_j+Q_j}\right|---\left(9\right)$

Wherein, u _iand S _irepresent mean value and the standard deviation of the i-th width test pattern evaluation score respectively; Jth name subject is classified to the judge mark of all test patterns, obtains the measuring parameter P of judge two class situation _jand Q _j.