CN108366295B - Video classification feature extraction method, transcoding recompression detection method and storage medium - Google Patents

Abstract

The present invention provides a video classification feature extraction method, a transcoding recompression detection method and a storage medium. The video classification feature extraction method includes: using a visual analyzer to extract the PU division type of a video frame, and extracting the extracted The PU division type of the video frame is marked with the pixel block as the basic unit; the number of pixel blocks corresponding to each PU division type of the first P frame in each group of consecutive pictures in the video is counted; The number of pixel blocks corresponding to each PU division type of the frame is averaged, and the classification features of each PU division type of the first P frame in all groups of consecutive pictures are obtained. The present invention uses a smaller number of dimensional features to achieve a higher detection rate of recompressed video.

Description

Video classification feature extraction method, transcoding recompression detection method and storage medium

technical field

The present invention relates to the technical field of video processing, and in particular, to a video classification feature extraction method, a video transcoding recompression detection method and a computer-readable storage medium.

Background technique

With the rapid development of the current Internet, the acquisition and transmission of digital video has become increasingly mature and popular. At the same time, increasingly powerful video editing software has become more and more popular, which makes digital video tampering happen from time to time. When the tampered video is used in judicial, media and other industries, it will distort the truth, which will lead to serious consequences of judicial misjudgment and media misrepresentation. Therefore, the authenticity and integrity of digital video has become an urgent problem to be solved in the current society.

At present, the research at home and abroad mainly focuses on digital images, and a lot of research results have also been obtained. For example, detecting illegal copies of copyrighted images, detection of copy movement in images, and distinguishing computer-generated images from photographic images. Due to the large amount of video information and various tampering methods, video forensics research is difficult. Relying on the progress of image forensics research, video forensics technology has also made great progress in recent years.

Common video tampering methods generally need to undergo tampering operations such as decoding, deleting frames or inserting frames. The tampered video sequence needs to be compressed again to regenerate the video stream. Therefore, detecting whether the video is recompressed can be used as a technical means to detect whether the video has been tampered with. In the prior art, there are many methods for recompression detection. For example, when recompressing an MPEG video at the same bit rate, the number of DCT coefficient changes in single-compression and double-compression is larger than that in double-compression and triple-compression. The number of changes is large, and this phenomenon is used for video recompression detection. The recompression of MPEG video is detected by examining the regularity of the blockiness intensity of the recompressed video and the change of its mean value. The difference of the motion compensation side effects of adjacent P frames is exploited, and the recompression of MPEG video is detected by judging whether there is a spike in the Fourier transform domain. Distinguish MPEG-2 single-shot compressed video and recompressed video by using the convexity of the statistical histogram of quantized DCT coefficients in recompressed video. The algorithm is suitable for detecting recompression using different MPEG-2 encoders and is robust to frame deletion and tampering. . A detection algorithm characterized by the probability of non-zero quantized AC coefficients is used to distinguish H.264 single-compressed video and double-compressed video, and a high classification accuracy is achieved when the quantization parameter of the second compression is smaller than the quantization parameter of the first compression. Rate. A multi-compression forensics algorithm for H.264 video under the same quantization parameters, constructs a feature set containing quartiles based on the ratio difference of different quantized DCT coefficients between adjacent three compressions, as the input of the support vector machine, realizes the Classification of single-compressed video and multi-compressed video. The proposed algorithm has high classification accuracy and strong robustness to copy/paste attack and frame deletion attack.

As the latest video coding standard, HEVC has attracted considerable attention of researchers. Compared with H.264, under the same video quality, HEVC provides double the data compression ratio, that is, under the same bit rate, HEVC can greatly improve the video quality. It supports resolutions up to 8192×4320, which also includes 8k UHD. Therefore, in the prior art, a method for recompression and measurement of HEVC video is proposed, for example, an HEVC recompression video detection algorithm based on the statistical characteristics of adjacent DCT coefficients on parity combination. At the same time, a co-occurrence matrix based on quantized DCT coefficients and an HEVC video recompression detection algorithm based on Markov feature optimization are proposed in the prior art. Taking advantage of the sudden change in the number of blocks corresponding to each PU division type of I frame in HEVC recompressed video, a forensic algorithm for HEVC video recompression at different bit rates is proposed. However, in the prior art methods for HEVC video recompression detection, the classification feature dimensions are large, the amount of calculation is large, and a high detection rate is not achieved.

Therefore, in order to solve the above-mentioned technical problems, there is a need for a video classification feature extraction method and a video transcoding recompression detection method with few classification feature dimensions and high detection rate.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a video classification feature extraction method and a video transcoding recompression detection method to solve at least one defect in the prior art.

One aspect of the present invention provides a video transcoding recompression detection method, the method includes:

Extract the prediction unit (PU) division type of the video frame by using a visual analyzer, and mark the extracted PU division type of the video frame with pixel blocks as the basic unit;

Count the number of pixel blocks corresponding to each PU division type of the first P frame in each group of consecutive pictures in the video;

The number of pixel blocks corresponding to each PU division type of the first P frame in all groups of consecutive pictures is averaged, and the classification features of each PU division type of the first P frame in all groups of consecutive pictures are obtained.

Preferably, when extracting the PU partition type of the video frame, the RGB component of the border color of the visual analyzer is set, preferably (255, 0, 255) or any other suitable value that can distinguish the PU partition border from the video content. component value.

Preferably, the PU division type of the video frame is marked with an N×N pixel block as a basic unit, where N is 4 or an integer multiple of 4. For example, an 8×8 pixel block can be used as a basic unit for marking.

Preferably, the average value of the number of pixel blocks corresponding to each PU division type of the first P frame in all groups of consecutive pictures is achieved by the following formula:

其中P_i＝{p_i,0,p_i,1,...，p_i,24}(i＝1，2，...，M)，M为视频中包含的连续画面的组数。

Wherein P _i ={pi _,0 ,pi _,1 ,...,pi _,24 }(i=1,2,...,M), where M is the number of groups of consecutive pictures included in the video.

Another aspect of the present invention is to provide a video transcoding recompression detection method, comprising:

Randomly select the same number of single-shot compressed videos and re-compressed videos as training samples and send them to the support vector machine;

Perform video classification feature extraction on the single compressed video and the recompressed video according to the following method: analyze the PU division type, use a visual analyzer to extract the PU division type of the video frame, and divide the extracted PU of the video frame into The type is marked with pixel blocks as the basic unit; count the number of pixel blocks corresponding to each PU division type of the first P frame in each group of consecutive pictures in the video; divide each PU of the first P frame in all groups of consecutive pictures The number of pixel blocks corresponding to the type is averaged, and the classification features of each PU division type of the first P frame in all groups of consecutive pictures are obtained;

The support vector machine constructs a decision function according to the extracted classification feature;

Randomly select the single-shot compression video and the re-compressed video for testing and send them to the support vector machine as test samples, and the support vector machine determines whether the test video is a single-shot compressed video or a re-compressed video according to the output of the decision function. Classification detection results.

其中，AR为评估指标，TNR为判定为单次压缩视频的比率；TPR为判定为重压缩视频的比率。Preferably, the method further includes calculating an evaluation index representing the classification performance in the following manner:

Among them, AR is the evaluation index, TNR is the ratio of the video that is judged as single compression; TPR is the ratio of the video that is judged to be re-compressed.

其中，n为测试和训练不同的视频样本的次数。Preferably, the method further comprises: calculating a detection rate representing video compression detection by:

where n is the number of times to test and train different video samples.

The recompressed video is the original video and is processed in any one of the standard formats of H.261, H.263, H.263+, H.264, MPEG-1, MPEG-2 and MPEG-4 at the first bit rate. compression. For example, the recompressed video is a video obtained by performing H.264 compression on the original video at a first bit rate, and then performing HEVC compression on the decoded video at a second bit rate after decoding.

Preferably, the single-shot compressed video is a video obtained by performing HEVC compression on the original video at the second bit rate.

The PU division type of the video frame is marked with an N×N pixel block as a basic unit, where N is 4 or an integer multiple of 4. For example, when extracting the PU division type of the video frame, the RGB component of the border color of the visual analyzer is selected as (255, 0, 255); the PU division type of the video frame is marked with an 8×8 pixel block as the basic unit .

Preferably, the average value of the number of pixel blocks corresponding to each PU division type of the first P frame in all groups of consecutive pictures is achieved by the following method:

其中P_i＝{p_i,0,p_i,1,...，p_i,24}(i＝1，2，...，M)，M为视频中包含的连续画面的组数。

Wherein P _i ={pi _,0 ,pi _,1 ,...,pi _,24 }(i=1,2,...,M), where M is the number of groups of consecutive pictures included in the video.

Yet another aspect of the present invention also provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed, the above-mentioned method steps are implemented.

The invention provides a video classification feature extraction method and a video transcoding recompression detection method, which perform recompression detection for video transcoding recompressed videos of H.264 to HEVC standards, and the extracted classification features have fewer dimensions, which can achieve higher high detection rate. In addition, the method provided by the present invention can also be applied to detect the video encoding standard before the first video encoding standard is other HEVC standard, for example, it can also be applied to H.261, H.263, H.263+, MPEG-1 , MPEG-2 and MPEG-4.

It is to be understood that both the foregoing general description and the following detailed description are exemplary illustrations and explanations, and should not be used as limitations on what is claimed in the present invention.

Description of drawings

Further objects, functions and advantages of the present invention will be elucidated by the following description of embodiments of the present invention with reference to the accompanying drawings, wherein:

FIG. 1 is a schematic diagram of a PU division mode under different prediction modes;

Fig. 2 is the flow chart of the video classification feature extraction method of the present invention;

3a to 3d are schematic diagrams of P-frame PU division types of single-shot compressed video and recompressed video;

FIG. 4 is a schematic diagram of marking of PU division types of the present invention;

Fig. 5 is a flow chart of the video transcoding and recompression detection method of the present invention.

Detailed ways

Objects and functions of the present invention and methods for achieving these objects and functions will be elucidated by referring to the exemplary embodiments. However, the present invention is not limited to the exemplary embodiments disclosed below; it may be implemented in various forms. The essence of the description is merely to assist those skilled in the relevant art to comprehensively understand the specific details of the present invention.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the drawings, the same reference numbers represent the same or similar parts, or the same or similar steps. The content of the present invention will be described below through specific embodiments. After the video tamperer performs frame deletion, insertion, etc. tampering on the video, he needs another video compression format to recompress the video sequence. Compared with other encoding standards, H.264 encoding is most related to HEVC encoding, and the encoding architecture is roughly similar. After tampering with the original video in H.264 format, HEVC encoding is used for recompression when recompressing. In this example, the H.264 coding standard is used as the standard for the first video compression.

After the video is transcoded and recompressed, due to the change of the pixel value of the key frame, when the first P frame in a group of continuous pictures (GOP) uses the key frame as the reference frame for inter-frame prediction encoding, the intra-frame PU division type also changes. will change accordingly. Based on this phenomenon, the present invention uses a histogram to count the number of pixel blocks corresponding to each PU division type in such a frame, and uses this as a classification feature to detect video transcoding and recompression.

In the embodiment, in order to clarify the transcoding and recompression detection of the present invention for other coding standards to HEVC, it is necessary to describe the PU division type of the P frame. HEVC coding adopts a hybrid coding framework of prediction and transform. Three basic units are introduced in HEVC coding, namely coding unit (CU), prediction unit (PU) and transform unit (TU), thus making HEVC more flexible than H.264/AVC coding. Among them, CU is the coding basic unit, PU is used for intra and inter prediction, and TU is used for transform and quantization. A PU is a basic unit that contains prediction information. A CU can be divided into one or more PUs. The PU prediction modes can be divided into skip, intra and inter. Figure 1 shows a schematic diagram of PU division types in different modes. When the prediction mode is skip mode a, the PU size can only be 2N×2N; when the prediction mode is intra mode b, the PU has 2N×2N and N×N two division modes; when the prediction mode is inter frame mode c, the PU has 8 division modes, including 2N×2N and N×N two square division modes c1, 2N×N and N×2N two symmetrical The division mode c2, and four asymmetric division modes c3 of 2N×nU, 2N×nD, nL×2N and nR×2N. The asymmetric split mode c3 is an optional mode, which can be turned on or off through the relevant syntax in the encoding configuration file.

In the transcoded video recompression process, the prediction information contained in the PU block explains the prediction process of the CU block. Figure 2 is a flowchart of a method for extracting video classification features according to the present invention. As shown in Figure 2, a method for extracting video classification features provided by an embodiment of the present invention includes the following steps:

Step S101, analyze the PU division type, extract the PU division type of the video frame and mark it.

For the analysis of the PU division type, as shown in Figures 3a to 3d are schematic diagrams of the P-frame PU division type of the single compressed video and the H.264 to HEVC recompressed video. In the embodiment, the PU division type is analyzed to extract the video's Categorical features. Figure 3a shows the PU partition type of the first P frame in the first group of continuous pictures (GOP) of bridge_far compressed at a bit rate of 3.5M using HEVC encoding. Figure 3b shows the PU partition type of the first P frame in the first group of consecutive pictures (GOPs) after compression at a bit rate of 3M using H.264 encoding and compression at a bit rate of 3.5M using HEVC encoding. Figure 3c shows the PU partition type of the first P frame in the first group of continuous pictures (GOP) of bridge_far after being compressed at 3.5M bit rate using H.264 encoding and then using HEVC encoding to compress at 3.5M bit rate. Figure 3d shows the PU partition type of the first P frame in the first group of continuous pictures (GOP) in the first group of continuous pictures (GOP) of bridge_far after using H.264 encoding at 4M bit rate and then using HEVC encoding to compress at 3.5M bit rate.

Table 1. Number of pixel blocks corresponding to each PU partition type

It can be seen from Table 1 that the distribution trend of the number of pixel blocks corresponding to each PU division type of 3M-3.5M, 3.5M-3.5M and 4M-3.5 is roughly the same, and the distribution of the number of pixel blocks corresponding to the 3.5M PU division type Big difference. Especially when the PU division types are 4×4, 8×8, 4×8, and 8×4, the number of pixel blocks corresponding to each PU division type of 3M-3.5M, 3.5M-3.5M and 4M-3.5 is the same as that of 3.5M. The distribution of the number of pixel blocks corresponding to the PU partition type has a large difference.

According to the above analysis of PU partition type analysis and Table 1, after using H.264 encoding, the PU partition type of the first P frame in the first group of continuous pictures (GOP) compressed by HEVC encoding is mainly small pixel blocks. ; The PU division type of the first P frame in the first group of continuous pictures (GOP) compressed by HEVC encoding at a time is dominated by large pixel blocks. Therefore, after using H.264 encoding, the PU of the first P frame in the first group of consecutive pictures (GOPs) compressed by HEVC encoding is a fine PU division type.

It should be understood that during H.264 coding and compression, DCT transform is performed on each pixel block separately, and the correlation between blocks is ignored. The pixel values at the block-to-block boundary appear discontinuous jumps. When the video is compressed by HEVC encoding, due to the influence of block effect, a smaller PU partition type is required at the boundary between blocks to express the image transition here. In addition to H.264, other pre-HEVC video coding standards also use block DCT transform, and there is also block effect. Therefore, it is foreseeable that after video compression is performed using pre-HEVC video coding standards, and then transcoded to HEVC format video, the Both require smaller PU partition types to express image transitions at block boundaries.

According to the present invention, the embodiment uses a visual analyzer to extract the PU division type of the video frame after analyzing the PU division type, and marks the PU division type of the extracted video frame with pixel blocks as the basic unit.

Specifically, in the embodiment, the visual analyzer may use video analysis software such as Gitl_HEVC_Analyze or any other suitable software to perform PU division type extraction on the video frame. To distinguish the video background from the PU partition type boundary, the RGB components of the visual analyzer boundary color are set to (255, 0, 255).

According to the present invention, in the embodiment, the PU division type of a video frame is marked with an 8×8 pixel block as the basic unit, and in the embodiment, the PU division type is marked in the form of a label. In the present invention, there are altogether 25 PU division types in the embodiment. As shown in FIG. 4 , a schematic diagram of marking of PU division types of the present invention is shown.

Table 2. PU partition types corresponding to labels

Step S102, count the number of pixel blocks corresponding to each PU division type of the first P frame in each group of consecutive pictures in the video.

The number of pixel blocks corresponding to each PU division type of the first P frame in each group of consecutive pictures in the video is counted. The number of pixel blocks corresponding to each PU division type of the first P frame in each group of consecutive pictures is recorded as: P _i ={pi _,0 ,pi _,1 ,...,pi _,24 }(i =1, 2, ..., M), M is the number of groups of consecutive pictures included in the video, that is, the number of 8×8 pixel blocks corresponding to 25 PU division types is recorded in each _Pi .

Step S103, extracting classification features of the video.

The number of pixel blocks corresponding to each PU division type of the first P frame in all groups of consecutive pictures is averaged, and the classification features of each PU division type of the first P frame in all groups of consecutive pictures are obtained. The average value of the number of pixel blocks corresponding to each PU division type of the first P frame in all groups of consecutive pictures is achieved by the following formula:

其中P_i＝{p_i,0,p_i,1,...，p_i,24}(i＝1，2，...，M)，M为视频中包含的连续画面的组数。

Wherein P _i ={pi _,0 ,pi _,1 ,...,pi _,24 }(i=1,2,...,M), where M is the number of groups of consecutive pictures included in the video.

Taking the average value of the number of pixel blocks corresponding to each PU division type of the first P frame in all groups of consecutive pictures as the histogram of the number of pixel blocks corresponding to each PU division type of the first P frame in each group of consecutive pictures, Get the classification features of the video.

The video transcoding recompression detection is performed by a method for extracting video classification features of the present invention, as shown in FIG.

Step S201, randomly select the same number of single-shot compressed videos and re-compressed videos as training samples and send them to the support vector machine.

The recompressed video is a video obtained by performing H.264 compression on the original video at the first bit rate, and then performing HEVC compression on the decoded video at the second bit rate after decoding. The single-shot compressed video is a video obtained by performing HEVC compression on the original video at the second bit rate.

Specifically, in the embodiment, the single-shot compressed video and the re-compressed video are first made as detection targets.

34 uncompressed YUV sequences are used as initial videos, including 17 QCIF format videos (with a resolution of 176×144) and 17 CIF format videos (with a resolution of 352×288). To increase the sample capacity, each video is segmented into non-overlapping video segments of length 100 frames. Finally, a total of 36 QCIF video clips and 43 CIF video clips are generated.

HM10.0 uses the encoder_lowdelay_P_main configuration file for the HEVC encoding and decoding process. JM uses the encoder_main configuration file for the H.264 encoding and decoding process. The frame rate, I-frame period, and GOP size are set to 30, 4, and 4, respectively.

Making a single-shot compressed video: The original video is obtained by HEVC compression at the second bit rate (B ₂ ).

Making a recompressed video: perform H.264 compression on the original video at the first bit rate (B ₁ ), and then perform HEVC compression on the decoded video at the second bit rate (B ₂ ) after decoding.

In the embodiment, since the QCIF and CIF videos have different spatial resolutions, different bit rates should be selected to ensure the visual quality of the encoded videos. For QCIF video, the values of the first bit rate (B ₁ ) and the second bit rate (B ₂ ) are selected from {100, 200, 300} (kbps) and {200, 300, 400} (kbps), respectively. For CIF video, the first bit rate (B ₁ ) and the second bit rate (B ₂ ) are selected from {3, 3.5, 4} (Mbps) and {3.5, 4, 4.5} (Mbps), respectively.

Among the single-shot compressed videos and re-compressed videos produced above, the same number of single-shot compressed videos and re-compressed videos are randomly selected as training samples and sent to a support vector machine (SVM) for training. In this embodiment, 30 single-compressed videos and 30 re-compressed videos are randomly selected for training for videos in QCIF format. For the videos in CIF format, 35 single-shot compressed videos and 35 re-compressed videos are randomly selected for training.

In the training phase, the support vector machine (SVM) performs steps 2 and 3 to construct a decision function.

Step 202: Extract classification features from the single-compressed video and the re-compressed video.

According to the present invention, the embodiment performs video classification feature extraction on single compressed video and recompressed video according to the following method:

Utilize the visual analyzer to extract the PU division type of the video frame, and mark the extracted PU division type of the video frame with pixel blocks as the basic unit;

Count the number of pixel blocks corresponding to each PU division type of the first P frame in each group of consecutive pictures in the video;

The number of pixel blocks corresponding to each PU division type of the first P frame in all groups of consecutive pictures is averaged, and the classification features of each PU division type of the first P frame in all groups of consecutive pictures are obtained.

Step S203, the support vector machine constructs a decision function according to the extracted classification features.

The extraction of the video classification features in step 2 has been explained in detail above, and will not be repeated here.

Preferably, in the embodiment, the support vector machine may select the LIBSVM open source software with the SVMcg kernel or other software with similar functions as the classifier.

Step S204 , randomly select the single-shot compressed video and the re-compressed video for testing as test samples and send them to the support vector machine, and output the classification result.

Randomly select the single-shot compression video and the re-compressed video for testing and send them to the support vector machine as test samples, and the support vector machine determines whether the test video is a single-shot compressed video or a re-compressed video according to the output of the decision function. Classification detection results.

According to the present invention, the present embodiment can calculate the evaluation index representing the classification performance in the following manner:

其中，AR为评估指标，TNR为判定为单次压缩视频的比率；TPR为判定为重压缩视频的比率。

Among them, AR is the evaluation index, TNR is the ratio of the video that is judged as single compression; TPR is the ratio of the video that is judged to be re-compressed.

The detection rate of video compression detection is calculated and expressed as follows:

其中，n为测试和训练不同的视频样本的次数。

where n is the number of times to test and train different video samples.

In this embodiment, 20 times of training and testing are selected to obtain the average value of the detection rate, and the average value of the detection rate of the specific video compression detection is calculated as follows:

其中，AR为评估指标，n＝20。

Among them, AR is the evaluation index, n=20.

In order to more clearly reflect the advantages of the video classification feature extraction method and the video transcoding recompression detection method provided by the present invention, the video classification feature extraction method and the video transcoding recompression detection method of the present invention are respectively adopted in this embodiment. It is compared with the video transcoding recompression detection method that uses the co-occurrence matrix of the number of pixel blocks corresponding to the PU division type of I frame as the video classification feature. Table 3 is the detection rate of the recompressed video of the QCIF format of the present invention, Table 3 is the detection rate of the recompressed video of the CIF format of the present invention, and Table 5 is the co-occurrence matrix of the number of pixel blocks corresponding to the PU division type of the I frame as the video Detection rate of recompressed video in QCIF format for video transcoding recompression detection method of classification features.

Table 3. Detection rate of the recompressed video of the QCIF format of the present invention

B₁/B₂ 200k 300k 400k 100k 0.9667 0.9167 0.9125 200k 0.9208 0.9750 0.9750 300k 0.9208 0.9417 0.9500

Table 4. Detection rate of recompressed video in CIF format of the present invention

B₁/B₂ 3.5M 4M 4.5M 3M 0.9813 0.9781 0.9875 3.5M 0.9875 0.9750 0.9688 4M 0.9813 0.9844 0.9813

Table 5. The detection rate of the recompressed video in the QCIF format using the co-occurrence matrix of the number of pixel blocks corresponding to the PU division type of the I frame as the video classification feature of the video transcoding recompression detection method

B₁/B₂ 200k 300k 400k 100k 0.7750 0.8417 0.8667 200k 0.8375 0.8709 0.8667 300k 0.7957 0.8375 0.8917

As can be seen from Table 3 and Table 4, using the video classification feature extraction method and the video transcoding recompression detection method of the present invention, the recompression detection rates of QCIF and CIF formats both reach more than 90%, and the highest reaches 98.75%, The lowest is 92.08%.

As can be seen from Table 5, in the prior art, the co-occurrence matrix of the number of pixel blocks corresponding to the PU division type of I frame is adopted as the video transcoding recompression detection method of the video classification feature, and the recompression video detection rate of the QCIF format is 77%. %-90%. The detection accuracy rate of the recompressed video of the present invention is between 91% and 97.5%, which is obviously higher than that of the prior art. Meanwhile, the video classification feature extraction method and the video transcoding recompression detection method of the invention are adopted in the embodiment, the PU division types are 25, and the co-occurrence matrix of the number of pixel blocks corresponding to the PU division type of one frame is used as the video in the prior art. There are 100 types of PU division in the video transcoding and recompression detection method of categorical features. The PU division type of the present invention is 1/4 of the prior art. In the present invention, the dimension of the PU division type is lower than that of the prior art, and at the same time, the calculation amount is reduced, and the detection rate of the recompressed video is improved. The video classification feature extraction method and the video transcoding and recompression detection method of the present invention are more effective.

The invention provides a video classification feature extraction method and a video transcoding recompression detection method, which perform recompression detection for video transcoding recompressed videos of H.264 to HEVC standards, and the extracted classification features have fewer dimensions, which can achieve higher high detection rate.

Portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one of the following techniques known in the art, or a combination thereof: having logic gates for implementing logic functions on data signals Circuits are discrete logic circuits, application specific integrated circuits with suitable combinational logic gates, Programmable Gate Arrays (PGA), Field Programmable Gate Arrays (FPGA), etc.

Logic and/or steps represented in flowcharts or otherwise described herein, for example, may be considered an ordered listing of executable instructions for implementing the logical functions, and may be embodied in any computer-readable storage medium , for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device), device or equipment. A "computer-readable storage medium" may include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like.

Features described and/or illustrated above for one embodiment may be used in the same or similar manner in one or more other embodiments, and/or in combination with or instead of features in other embodiments features are used.

It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps may be performed in the order mentioned in the embodiments, or may be different from the order in the embodiments, or several steps may be performed simultaneously.

In addition, the method provided by the present invention can also be applied to other scenarios where the compression format standard before the HEVC standard is transcoded into the HEVC format, for example, it can also be applied to H.261, H.263, H.263+, MPEG-1, MPEG -2 and either standard MPEG-4.

Other embodiments of the present invention will be readily apparent to and understood by those skilled in the art in conjunction with the specification and practice of the present invention disclosed herein. The description and examples are to be regarded as exemplary only, with the true scope and spirit of the invention being defined by the claims.

Claims (9)

1. A video classification feature extraction method is characterized by comprising the following steps:

extracting a PU partition type of a prediction unit of a video frame by using a visual analyzer, and marking the extracted PU partition type of the video frame by using a pixel block as a basic unit;

counting the number of pixel blocks corresponding to each PU partition type of a first P frame in each group of continuous pictures in the video;

and calculating the average value of the pixel block numbers corresponding to the PU partition types of the first P frame in all the groups of continuous pictures to obtain the classification characteristics of the PU partition types of the first P frame in all the groups of continuous pictures.

2. The method of claim 1, wherein the PU partition type of the video frame is marked with a basic unit of N × N pixel blocks, wherein N is 4 or 8.

3. The method according to claim 1, wherein the averaging of the number of pixel blocks corresponding to each PU partition type of the first P frame in all groups of consecutive pictures is performed by the following formula:

wherein P is_i＝{p_i,0,p_i,1,...，p_i,24M is the number of groups of consecutive pictures contained in the video.

4. A method for detecting transcoding and recompression of a video, the method comprising:

randomly selecting single compressed videos and re-compressed videos with the same number as training samples to be sent to a support vector machine;

performing video classification feature extraction on the single-pass compressed video and the re-compressed video according to the video classification feature extraction method of any one of claims 1 to 3;

the support vector machine constructs a decision function according to the extracted classification features;

and randomly selecting a single compressed video and a recompressed video for testing as test samples and sending the test samples to the support vector machine, wherein the support vector machine outputs a classification detection result for judging whether the tested video is the single compressed video or the recompressed video according to the decision function.

5. The method of claim 4, further comprising:

an evaluation index representing classification performance is calculated by:

wherein, AR is an evaluation index, and TNR is a ratio judged as a single-pass compressed video; the TPR is a ratio determined to recompress the video.

6. The method of claim 5, further comprising:

the detection rate representing the detection of video compression is calculated by:

where n is the number of times different video samples are tested and trained.

7. The method according to claim 4, wherein the re-compressed video is an original video compressed in any one of standard formats of H.261, H.263+, H.264, MPEG-1, MPEG-2 and MPEG-4 at a first bit rate, and the original video is subjected to HEVC compression at a second bit rate after being decoded.

8. The method of claim 4, wherein the single-pass compressed video is HEVC compressed video at a second bit rate from an original video.

9. A computer-readable storage medium, in which a computer program is stored which, when being executed, carries out the method steps of any one of claims 1 to 8.

Images