CN103327320A - Identification method used for fake high code rate video - Google Patents
Identification method used for fake high code rate video Download PDFInfo
- Publication number
- CN103327320A CN103327320A CN201310102134XA CN201310102134A CN103327320A CN 103327320 A CN103327320 A CN 103327320A CN 201310102134X A CN201310102134X A CN 201310102134XA CN 201310102134 A CN201310102134 A CN 201310102134A CN 103327320 A CN103327320 A CN 103327320A
- Authority
- CN
- China
- Prior art keywords
- video
- code check
- code rate
- identified
- quality
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 19
- 238000012706 support-vector machine Methods 0.000 claims abstract description 6
- 230000008859 change Effects 0.000 claims description 20
- 238000012360 testing method Methods 0.000 claims description 6
- 239000004744 fabric Substances 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 3
- 230000007423 decrease Effects 0.000 abstract 2
- 230000003247 decreasing effect Effects 0.000 abstract 1
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012797 qualification Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010396 two-hybrid screening Methods 0.000 description 1
Images
Abstract
The invention discloses an identification method used for a fake high code rate video, and belongs to the field of safety of multi-media information. The identification method is used for identifying the original code rate (bit rate) of the fake code rate video and is a statistic judging method put forward according to characteristics of correlation between the video code rate and objective quality. According to the characteristic that the objective quality of the fake code rate video does not decrease in a monotonic mode when the code rate is decreased, a characteristic curve is extracted, and the original code rate of the video which undergoes upconversion through the code rate can be effectively detected. According to the implementation method of the method, a video to be identified is recoded, a series of recoded videos in low code rate version are acquired, quality difference between each recoded video and the video to be identified is calculated, the changing curve in which the quality decreases along with the code rate is drawn, three kinds of matching is performed on the quality-code rate changing curve, three matching determination coefficients are extracted and used as three-dimensional characteristics of the video to be identified, and the three-dimensional characteristics are input into a support vector machine classifier to be trained and tested.
Description
Technical field
The present invention relates to the multi-media information security field, more specifically, relate to a kind of authentication method for the high bit-rate video of puppet.
Background technology
The develop rapidly of science and technology makes obtaining of multimedia video become more and more easier.It is popular that intelligence shooting mobile phone and video camera have promoted that especially video is autodyned and shared with video.On general video sharing website, the quality of video is often classified by the code check of video, and the video of high code rate is marked as HD video, and these videos also more are subjected to online friend's favor usually.The online friend in order to obtain higher clicking rate, may improve video code rate by switching software before being uploaded to the video sharing website, thus the pseudo-video of producing high code check.The high bit-rate video that these puppets are produced, video quality does not get a promotion, and the high bit-rate video that claims this type of puppet to produce is " pseudo-high bit-rate video ".Yet whether the code check for video passes through forgery, and the reliable technological means of Shang Weiyou is identified.
Summary of the invention
In order to overcome the deficiencies in the prior art, the present invention has provided a kind of authentication method for the high bit-rate video of puppet, can identify video reliably and whether live through the code check conversion, can estimate original code check simultaneously.
For solving the problems of the technologies described above, technical scheme of the present invention is as follows:
A kind of authentication method for the high bit-rate video of puppet may further comprise the steps:
S1. recompile video to be identified obtains the recodification video of a series of low code check versions;
S2. calculate the mass change between each recodification video and the video to be identified;
S3. draw out the change curve of quality-recodification code check;
S4. according to the change curve of quality-recodification code check, adopt three kinds of curve model matches to carry out;
S5. extract the match coefficient of determination of above-mentioned three curve models, with the three-dimensional feature of three match coefficients of determination as video to be identified, be input to SVMs svm classifier device and carry out training and testing.
When the change curve to quality-recodification code check carries out match, what kind of curve model the curve of considering quality-recodification code check meets, in fact more match not necessarily obtains more accurate result, has chosen three kinds of curve models that are used for match in the present invention.
Preferably, utilize the video coding tools FFmpeg that increases income that video v to be identified is carried out recompile among the described step S1, the code check that makes video v to be identified is bKbps, reduces cbr (constant bit rate) at every turn and is respectively: b
1, b
2, b
3..., b
k, b wherein
I+1-b
i=b
i-b
I-1, i=2...k-1 is encoded into a series of low code check version videos, is respectively: v
1, v
2, v
3..., v
k
Its step-length that reduces code check of video for different resolution is also different, and for example the step-length of 32Kbps is applicable to the video of resolution 352x288 pixel.Little step-length can obtain accurate more result, but the computation complexity that thereupon increases is huge.
Preferably, utilize SSIM to calculate recodification video v among the described step S2
i, the mass change between i=1...k and the video v to be identified is calculated respectively each frame, obtains the mean value S of its whole section sequence at last
i, mean value S
iNamely represent video v
iAnd the mass change between the video v to be identified, its specifically:
X wherein, y is two two field pictures that two videos align in time, wherein μ
xAnd μ
yBe two width of cloth image x, the average of y, σ
x, σ
yBe image corresponding variance, σ
XyBe image x, the covariance of y, c
1, c
2Be two constants, c
1=(k
1L)
2, c
2=(k
2L)
2, L=255 under the default situations, k
1=0.01, k
2=0.03.。
Preferably, among the described step S4 quality-recodification code check change curve is carried out three kinds of curve matches, three kinds of curve models are respectively: 2 order polynomials, 2 indexes of 1 exponential sum; Be specially: 2 order polynomials: f (x)=p
1x
2+ p
2X+p
3, wherein x represent to recode code check, f (x) represented quality, p
1, p
2, p
3For adopting the coefficient of correspondence of 2 order polynomials institute match; 1 index: f (x)=ae
Bx, wherein x represent to recode code check, f (x) represented quality, a, b are coefficient of correspondence and 2 index: f (the x)=ae of 1 index institute of employing match
Bx+ ce
Dx, wherein x represent to recode code check, f (x) represented quality, and a, b, c, d are the coefficient of correspondence of 2 index institutes of employing match, and wherein x represent to recode code check, f (x) represented quality.
Preferably, described step S5 match coefficient of determination r
k, k=1,2,3, r
k∈ [0,1].Match coefficient of determination r
kThe more big representative fitting degree of value more high.With the three-dimensional feature of three match coefficients of determination as video v to be identified, the grader that is input to SVMs SVM carries out training and testing.
Compared with prior art, the beneficial effect of technical solution of the present invention is:
The rule that video quality changed when the present invention reduced by analyzing video code rate judges whether whether video improves through code check, be original code check thereby identify current video, and identify the original code check of video by support vector machine classifier.The present invention can effectively detect the original code check of video that improves through code check, for the evaluation of the original code check of video provides effectively, simple method.
Description of drawings
Fig. 1 is the process that video code rate is distorted.
Fig. 2 is the flow chart that the present invention realizes.
Fig. 3 is the code check-figure-of-merit curve of recodification video.
Fig. 4 is three-dimensional feature instance graph (CIF form).
Embodiment
Below in conjunction with accompanying drawing the present invention is described further, but embodiments of the present invention are not limited to this.
Be illustrated in figure 2 as the flow chart of a kind of authentication method for the high bit-rate video of puppet of the present invention, it specifically comprises the steps:
(S1) video recompile to be identified is obtained a series of low code check versions; The utilization video coding tools FFmpeg that increases income carries out recompile to video v to be identified, supposes that its code check is bKbps, reduces cbr (constant bit rate) b respectively at every turn
1, b
2, b
3..., b
k, b wherein
I+1-b
i=b
i-b
I-1, i=2 ..., k-1 arrives a series of low code check version v with its code conversion
1, v
2, v
3..., v
kIts step-length that reduces code check of video for different resolution is also different, and for example the step-length of 32Kbps is applicable to the video of resolution 352x288 pixel, and step-length refers to b
I+1-b
i=b
i-b
I-1Tolerance.Little step-length can obtain accurate more result, but the computation complexity that thereupon increases is huge, so provide the reference step-length of table 1:
Table 1
| Resolution | 176x144 | 352x288 | 416x240 | 704x576 | 832x480 | 1280x720 |
| Step-length (bps) | 8K | 32K | 32K | 128K | 128K | 192K |
In the present embodiment, establish the code check b=1536Kbps of video v to be identified, resolution is 352x288, then reduces cbr (constant bit rate) 32Kbps at every turn, and the code conversion of video to be identified is respectively v to a series of low code check versions
1, v
2, v
3, v
4..., i.e. corresponding code check 1504,1472,1440,1408Kbps ...
(S2) calculate mass change between each recodification video and the video to be identified; Utilize SSIM to calculate recodification video v
i(i=1...k) and the mass change between the video v to be identified, two frames that align in two videos are calculated, obtain the mean value S of its whole section all frame of sequence at last
i, mean value S
iNamely represent recodification video v
iAnd the mass change between the video v to be identified, its specifically:
X wherein, y are two two field pictures of alignment, wherein μ in time in two videos
xAnd μ
yBe two width of cloth image x, the average of y, σ
x, σ
yBe two width of cloth image corresponding variance, σ
XyBe two width of cloth image x, the covariance of y, c
1, c
2Be two constants, c
1=(k
1L)
2, c
2=(k
2L)
2, L=255 under the default situations, k
1=0.01, k
2=0.03.
Calculate S respectively
1=Q (v
1, v), S
2=Q (v
2, v), S
3=Q (v
3, v), S
4=Q (v
4, v) etc.
(S3) draw out the change curve of quality-recodification code check;
Be illustrated in figure 3 as quality-recodification code check change curve of standard test sequences foreman, resolution is 352x288, article four, the video of curve representative all is 1536Kbps, what dotted line represented is the video of true code check, other three all have low code check to convert, are the original code check of each curve in the bracket.
(S4) the mass change curve is carried out three kinds of matches; Three kinds of curve models that quality-recodification rate curve carried out match are respectively: 2 order polynomials: f (x)=p
1x
2+ p
2X+p
3, wherein x represent to recode code check, f (x) represented quality, p
1, p
2, p
3For adopting the coefficient of correspondence of 2 order polynomials institute match; 1 index: f (x)=ae
Bx, wherein x represent to recode code check, f (x) represented quality, a, b are coefficient of correspondence and 2 index: f (the x)=ae of 1 index institute of employing match
Bx+ ce
Dx, wherein x represent to recode code check, f (x) represented quality, and a, b, c, d are the coefficient of correspondence of 2 index institutes of employing match, and wherein x represent to recode code check, f (x) represented quality.In the present embodiment, utilize the curve of MATLAB to carry out match, obtain above-mentioned three kinds of model of fit, and obtain the match coefficient of determination r of a model of fit
k, k=1,2,3.
(S5) with match coefficient of determination r
kAs three-dimensional feature input grader SVM, train; The match coefficient of determination r of three kinds of model of fit wherein
k, k=1,2,3, r
kGenerally in [0,1] scope, it is more high to be worth more big representative fitting degree.Three match coefficient of determination r
kAs the three-dimensional feature of video v to be identified, be input to SVMs svm classifier device and train and classify.In the present embodiment, import the video of half and train, the video of half is classified.
The three-dimensional feature that is illustrated in figure 4 as 95 CIF test video sequence distributes, and three reference axis are represented three match coefficient of determination r respectively
1, r
2, r
3It all is 1536Kbps that the code check of all videos " is observed ", and wherein the characteristic point represented of circle represents true bit-rate video, and its excess-three kind converts by low code check, the original code check of each curve in the bracket.As can be observed from Figure, the video of original code check and the video of pseudo-code check can divide, and be same, and it also is different that its feature of the video of different original code checks distributes, and just can divide.
Video database (95 sections CIF videos and 108 sections QCIF video sequences) for two kinds of resolution utilizes the present invention to carry out concrete experimental result shown in two hybrid matrix of following table:
Table 2
Table 3
Video to be identified is CIF form (352x288) in the above-mentioned table 2, wherein every group of experiment includes 95 sections videos, form first is classified the original code check of video as, first line display of form adopts institute of the present invention qualification result, table content is verification and measurement ratio (%), and is wherein alternative with * less than 5% verification and measurement ratio." observing " code check of all videos to be detected all is 1536Kbps, comprises true bit-rate video and pseudo-code check video, and wherein pseudo-code check video is converted by three kinds of code checks of 256,512,768Kbps.For each video to be identified, at first be that step-length reduces the recodification of code check to it with 32K, corresponding code check is respectively 1504,1472,1440,1408Kbps etc., calculate the mass value of each recodification video and this evaluation video then, comprehensive all recodification videos obtain quality-code check change curves, and next the match that curve is carried out three kinds of models obtains the match coefficient of determination under three kinds of models, the used three-dimensional feature of namely classifying.As above-mentioned step all videos to be identified are extracted three-dimensional feature, select half video features to be input in the svm classifier device at random and train, with remaining half carry out testing classification.Training-classification is carried out ten times altogether at random, and average ten times qualification result is namely shown in the form.
Video to be identified is QCIF form (176x144) in the above-mentioned table 3, and wherein every group of experiment includes 108 sections videos, and process is the same.
Can obviously find out from above data, utilize the video of the present invention to improving through code check, no matter be that CIF or QCIF form can both be predicted its original code check exactly, and verification and measurement ratio illustrates that the present invention has higher accuracy for the evaluation of the original code check of the high bit-rate video of puppet between 86.3%-94.44%.
Above-described embodiments of the present invention do not constitute the restriction to protection range of the present invention.Any modification of within spiritual principles of the present invention, having done, be equal to and replace and improvement etc., all should be included within the claim protection range of the present invention.
Claims (5)
1. an authentication method that is used for pseudo-high bit-rate video is characterized in that, may further comprise the steps:
S1. recompile video to be identified obtains the recodification video of a series of low code check versions;
S2. calculate the mass change between each recodification video and the video to be identified;
S3. draw out the change curve of quality-recodification code check;
S4. according to the change curve of quality-recodification code check, adopt three kinds of curve model matches to carry out;
S5. extract the match coefficient of determination of above-mentioned three curve models, with the three-dimensional feature of three match coefficients of determination as video to be identified, be input to SVMs svm classifier device and carry out training and testing.
2. the authentication method for the high bit-rate video of puppet according to claim 1, it is characterized in that, utilize the video coding tools FFmpeg that increases income that video v to be identified is carried out recompile among the described step S1, the code check that makes video v to be identified is bKbps, reduces cbr (constant bit rate) at every turn and is respectively: b
1, b
2, b
3..., b
k, b wherein
I+1-b
i=b
i-b
I-1, i=2 wherein ..., k-1; Be encoded into a series of low code check version videos, be respectively: v
1, v
2, v
3..., v
k
3. the authentication method for the high bit-rate video of puppet according to claim 1 is characterized in that, utilizes SSIM to calculate recodification video v among the described step S2
i, the mass change between i=1...k and the video v to be identified is calculated respectively each frame, obtains the mean value S of its whole section sequence at last
i, mean value S
iNamely represent video v
iAnd the mass change between the video v to be identified, its specifically:
X wherein, y is two two field pictures that two videos align in time, wherein μ
xAnd μ
yBe two width of cloth image x, the average of y, σ
x, σ
yBe image corresponding variance, σ
XyBe image x, the covariance of y, c
1, c
2Be two constants.
4. the authentication method for the high bit-rate video of puppet according to claim 1, it is characterized in that, among the described step S4 quality-recodification code check change curve is carried out three kinds of curve matches, three kinds of curve models are respectively: 2 order polynomials, 2 indexes of 1 exponential sum; Be specially: 2 order polynomials: f (x)=p
1x
2+ p
2X+p
3, wherein x represent to recode code check, f (x) represented quality, p
1, p
2, p
3For adopting the coefficient of correspondence of 2 order polynomials institute match; 1 index: f (x)=ae
Bx, wherein x represent to recode code check, f (x) represented quality, a, b are coefficient of correspondence and 2 index: f (the x)=ae of 1 index institute of employing match
Bx+ ce
Dx, wherein x represent to recode code check, f (x) represented quality, and a, b, c, d are the coefficient of correspondence of 2 index institutes of employing match, and wherein x represent to recode code check, f (x) represented quality.
5. the authentication method for the high bit-rate video of puppet according to claim 1 is characterized in that, the match coefficient of determination r of described step S5
k, k=1,2,3, r
k∈ [0,1].
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310102134.XA CN103327320B (en) | 2013-03-27 | 2013-03-27 | A kind of identification method used for fake high code rate video |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310102134.XA CN103327320B (en) | 2013-03-27 | 2013-03-27 | A kind of identification method used for fake high code rate video |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103327320A true CN103327320A (en) | 2013-09-25 |
| CN103327320B CN103327320B (en) | 2016-06-22 |
Family
ID=49195819
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310102134.XA Expired - Fee Related CN103327320B (en) | 2013-03-27 | 2013-03-27 | A kind of identification method used for fake high code rate video |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103327320B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123935A (en) * | 2014-07-16 | 2014-10-29 | 武汉大学 | Double compression detection method towards MP3 (moving picture experts group audio Layer-3) digital audio file |
| CN105376584A (en) * | 2015-11-20 | 2016-03-02 | 信阳师范学院 | Video motion compensation frame rate up-conversion evidence collection method based on noise level estimation |
| CN111723735A (en) * | 2020-06-18 | 2020-09-29 | 四川大学 | Pseudo high bit rate HEVC video detection method based on convolutional neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1816129A (en) * | 2005-02-01 | 2006-08-09 | 松下电器产业株式会社 | Recording apparatus |
| CN1914878A (en) * | 2004-02-12 | 2007-02-14 | 诺基亚公司 | Classified media quality of experience |
| CN101388977A (en) * | 2007-09-13 | 2009-03-18 | 株式会社东芝 | Image processing apparatus and image processing method |
| US20090074073A1 (en) * | 2003-07-18 | 2009-03-19 | Microsoft Corporation | Coding of motion vector information |
| CN101534436A (en) * | 2008-03-11 | 2009-09-16 | 深圳市融合视讯科技有限公司 | Allocation method of video image macro-block-level self-adaptive code-rates |
-
2013
- 2013-03-27 CN CN201310102134.XA patent/CN103327320B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090074073A1 (en) * | 2003-07-18 | 2009-03-19 | Microsoft Corporation | Coding of motion vector information |
| CN1914878A (en) * | 2004-02-12 | 2007-02-14 | 诺基亚公司 | Classified media quality of experience |
| CN1816129A (en) * | 2005-02-01 | 2006-08-09 | 松下电器产业株式会社 | Recording apparatus |
| CN101388977A (en) * | 2007-09-13 | 2009-03-18 | 株式会社东芝 | Image processing apparatus and image processing method |
| CN101534436A (en) * | 2008-03-11 | 2009-09-16 | 深圳市融合视讯科技有限公司 | Allocation method of video image macro-block-level self-adaptive code-rates |
Non-Patent Citations (2)
| Title |
|---|
| 徐俊瑜: "数字视频被动取证技术研究", 《中国优秀硕士学位论文全文数据库信息科技Ⅰ辑，2012年第03期》 * |
| 骆伟祺等: "鲁棒的区域复制图像篡改检测技术", 《计算机学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123935A (en) * | 2014-07-16 | 2014-10-29 | 武汉大学 | Double compression detection method towards MP3 (moving picture experts group audio Layer-3) digital audio file |
| CN104123935B (en) * | 2014-07-16 | 2017-02-22 | 武汉大学 | Double compression detection method towards MP3 (moving picture experts group audio Layer-3) digital audio file |
| CN105376584A (en) * | 2015-11-20 | 2016-03-02 | 信阳师范学院 | Video motion compensation frame rate up-conversion evidence collection method based on noise level estimation |
| CN105376584B (en) * | 2015-11-20 | 2018-02-16 | 信阳师范学院 | Turn evidence collecting method in video motion compensation frame per second based on noise level estimation |
| CN111723735A (en) * | 2020-06-18 | 2020-09-29 | 四川大学 | Pseudo high bit rate HEVC video detection method based on convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103327320B (en) | 2016-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Automatic detection of object-based forgery in advanced video | |
| CN102254006B (en) | Method for retrieving Internet video based on contents | |
| CN101835040B (en) | Digital video source evidence forensics method | |
| CN111160102B (en) | Training method of face anti-counterfeiting recognition model, face anti-counterfeiting recognition method and device | |
| CN105160678A (en) | Convolutional-neural-network-based reference-free three-dimensional image quality evaluation method | |
| CN109657600B (en) | Video area removal tampering detection method and device | |
| CN105426826A (en) | Tag noise correction based crowd-sourced tagging data quality improvement method | |
| CN104282310B (en) | Steganography detection method for audio subjected to MP3Stego steganography | |
| Kwon et al. | Learning self-similarity in space and time as generalized motion for video action recognition | |
| CN110457996B (en) | Video moving object tampering evidence obtaining method based on VGG-11 convolutional neural network | |
| CN104954807B (en) | The video dubbing altering detecting method of resist geometric attackses | |
| CN107396095A (en) | One kind is without with reference to three-dimensional image quality evaluation method | |
| Gu et al. | Structural similarity weighting for image quality assessment | |
| Fu et al. | Blind image quality assessment for multiply distorted images via convolutional neural networks | |
| CN103327320B (en) | A kind of identification method used for fake high code rate video | |
| CN110121109A (en) | Towards the real-time source tracing method of monitoring system digital video, city video monitoring system | |
| CN105120294A (en) | JPEG format image source identification method | |
| Castillo Camacho et al. | A simple and effective initialization of CNN for forensics of image processing operations | |
| CN104809248A (en) | Video fingerprint extraction and retrieval method | |
| Moussa et al. | Forensic license plate recognition with compression-informed transformers | |
| CN103413336B (en) | Detection method and the device of the dual JPEG compression of a kind of grid non-alignment | |
| CN105959685B (en) | A kind of compression bit rate Forecasting Methodology based on video content and cluster analysis | |
| Si et al. | A full‐reference stereoscopic image quality assessment index based on stable aggregation of monocular and binocular visual features | |
| Cemiloglu et al. | Blind video quality assessment via spatiotemporal statistical analysis of adaptive cube size 3D‐DCT coefficients | |
| Rodríguez-Lois et al. | A Critical Look into Quantization Table Generalization Capabilities of CNN-based Double JPEG Compression Detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160622 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |