CN104661035A - Compression method and system for local feature descriptor of video and video compression method - Google Patents

Abstract

The invention discloses a compression method and system for a local feature descriptor of a video and a video compression method. The compression method for the local feature descriptor of the video comprises the following steps: performing inter-frame prediction, quantization and entropy coding on the combined video content, calculating a residual coefficient of a prediction signal, quantizing the residual coefficient to obtain a quantization coefficient, performing entropy coding on the quantization coefficient, outputting a code stream formed after entropy coding, and finishing compression of the local feature descriptor of the video. The local feature descriptor of the video is compressed through the compression method and system, compressed video data is compactly expressed on the basis of the combined video content, and high compression ratio is achieved, so that the transmission rate of the video data is increased, the storage efficiency and the compression efficiency of the video data are improved, and the retrieval efficiency of the video data is improved.

Description

Video local feature descriptor compression method, system and video compression method

technical field

The invention belongs to the technical field of data processing, and in particular relates to a compression method, a compression system and a video compression method of a video local feature descriptor.

Background technique

With the wide application of the Internet, the data of video data is getting larger and larger in the process of network transmission and storage. Compression of video data is particularly important. Local feature descriptors have been widely used in image/video data processing and computer vision. Descriptor compression is also a key technology to improve video data applications such as video data retrieval. For example, for the 128-dimensional SIFT local feature descriptor, each dimension is quantized to 8 bits, and one frame of image contains 1000 SIFT feature points, then the amount of SIFT feature data contained in a 300-frame video is 300*1000* 128*8bits=292.97Mb, no matter for storage or transmission, it needs to pay a huge price, which is unbearable in practical applications.

In the prior art of video data processing, most of the descriptors are compressed separately, which limits the compression ratio of the descriptors, is not conducive to the improvement of data retrieval efficiency, and also hinders the development of video retrieval technology.

Contents of the invention

The purpose of the embodiment of the present invention is to provide a video local feature descriptor compression method, system and video compression method, so as to improve the compression efficiency of the video local feature descriptor, thereby improving the transmission rate and storage efficiency of video compressed data.

According to one aspect of the present invention, a method for compressing video local feature descriptors is provided, said method comprising the steps of:

Combine video content to perform inter-frame prediction on the current local feature descriptor to obtain a prediction signal;

calculating residual coefficients of said predicted signal;

Quantizing the residual coefficients to obtain quantized coefficients;

Entropy encoding is performed on the quantized coefficients to output a code stream.

In the above solution, after calculating the residual coefficient and before obtaining the quantized coefficient, the method further includes:

transforming the residual coefficients to obtain transformation coefficients;

Quantizing the residual coefficients to obtain quantized coefficients, and further, quantizing the transformation coefficients to obtain quantized coefficients.

In the above solution, the method further includes: performing inverse quantization and inverse transformation on the quantized coefficients to obtain a reconstruction descriptor, and storing the reconstruction descriptor.

In the above solution, the joint video content performs inter-frame prediction on the current local feature descriptor to obtain a prediction signal, which further includes: finding a frame that is identical to the current local feature descriptor in the coded previous frame of the frame where the current local feature descriptor is located The reconstruction descriptor closest to the current local feature descriptor is used as a prediction signal.

In the above solution, the quantization is scalar quantization or vector quantization.

According to another aspect of the present invention, a compression system of video local feature descriptors is also provided, the system includes: a prediction module, a residual coefficient calculation module, a quantization module, and an encoding module

The prediction module is used to perform inter-frame prediction on the current local feature descriptor in conjunction with video content to obtain a prediction signal;

The residual coefficient calculation module is connected to the prediction module, and is used to calculate the residual coefficient of the prediction signal;

The quantization module is connected to the residual coefficient calculation module, and is used to quantize the residual coefficient to obtain a quantized coefficient;

The coding module is connected to the quantization module, and is used for performing entropy coding on the quantization coefficients and outputting a code stream.

In the above solution, the system further includes a transformation module, the transformation module is connected to the residual coefficient calculation module and the quantization module, and is used to transform the residual coefficient to obtain a transformation coefficient;

The quantization module is also used to quantize the transform coefficients and obtain quantized coefficients.

In the above solution, the system further includes: a reconstruction descriptor storage module, configured to perform inverse quantization and inverse transformation on the quantized coefficients to obtain a reconstruction descriptor, and store the reconstruction descriptor.

In the above solution, the prediction module is further used to find a reconstructed descriptor closest to the current local feature descriptor in the coded previous frame of the frame where the current local feature descriptor is located, as a prediction Signal.

According to still another aspect of the present invention, a video compression method is also provided, the method comprising:

Compress the original video data to obtain the code stream of the video itself, and obtain the reconstructed video data composed of reconstructed frames;

In combination with the content of the reconstructed frame in the video reconstruction data, the local feature descriptor in the original video data is compressed by using the compression method of the video local feature descriptor described in any one of claims 1 to 5, to obtain Entropy coded descriptor stream;

Output the description sub-code stream into the video code stream itself, and output the video code stream.

Embodiments of the present invention provide a compression method, system, and video compression method for local video feature descriptors. The compression method for local video feature descriptors includes prediction, quantization, and entropy coding. Prediction, and then calculate the residual coefficient of the prediction signal, quantize the residual coefficient to obtain the quantized coefficient, and then perform entropy coding on the quantized coefficient, output the code stream, and complete the compression of the local feature descriptor of the video. Through the method of the embodiment of the present invention, the local feature descriptor of the video is compressed, and the compressed video data is compactly represented on the basis of joint video content, achieving a high compression ratio, thereby improving the transmission rate and storage of the video data Efficiency, while improving the retrieval efficiency of video data.

Description of drawings

Fig. 1 is a schematic flow chart of a method for compressing video local feature descriptors according to the first embodiment of the present invention;

Fig. 2 is a schematic flow chart of a method for compressing video local feature descriptors according to the second embodiment of the present invention;

Fig. 3 is a flowchart of a compression method in a specific example of the second embodiment of the present invention;

4 is a schematic structural diagram of a compression system of a video local feature descriptor according to a third embodiment of the present invention;

5 is a schematic structural diagram of a compression system of a video local feature descriptor according to a fourth embodiment of the present invention;

FIG. 6 is a schematic flowchart of a video compression method according to a fifth embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in combination with specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are exemplary only, and are not intended to limit the scope of the present invention. Also, in the following description, descriptions of well-known structures and techniques are omitted to avoid unnecessarily obscuring the concept of the present invention.

The embodiment of the present invention is based on the wide application of local feature descriptors in the field of computer vision, compresses video local feature descriptors, and applies the reconstructed video content itself with reconstructed descriptor content in the compression process, That is to combine video content, guide the compression of local feature descriptors, and propose a compression method, system and video compression method for video local feature descriptors. Hereinafter, the present invention will be described in detail in conjunction with the embodiments and the accompanying drawings.

FIG. 1 is a schematic flowchart of a method for compressing video local feature descriptors according to the first embodiment of the present invention. In this embodiment, the framework of the method for compressing video local feature descriptors includes joint prediction, quantization, and entropy coding of video content.

As shown in Figure 1, the compression method of the video local feature descriptor of the present embodiment comprises the following steps:

Step S101 , inter-frame prediction is performed on the current local feature descriptor in conjunction with the video content to obtain a prediction signal.

In this step, the inter-frame prediction is performed on the current local feature descriptor, specifically, in the encoded frame of the video, a reconstructed descriptor closest to the current local feature descriptor is found as predictive signal. Video data is composed of frames, and each frame has a corresponding local feature descriptor. The frame to which the current local feature descriptor in the process of compression belongs, that is, the current frame. Preferably, the encoded frame of the video is the previous frame of the current frame. The closest here means that the encoded reconstructed descriptor has the most identical feature data with the current descriptor, and it can also be selected through the sum of absolute value errors (SAD, Sum of Absolute Difference). This optimal value selection can be realized by a programmable logic controller (PLC). When searching for the closest reconstructed descriptor in the previous frame, you can search in all the data in the previous frame, or you can customize the search range, for example, search according to the motion vector of the video, in the current descriptor motion vector At the pointed position, the closest reconstruction descriptor is searched within the range of the five nearest nodes.

In particular, during the above-mentioned inter-frame prediction, the reconstruction descriptor that is the prediction result of the current local feature descriptor, in addition to the closest reconstruction descriptor within the corresponding range, can also be defined according to other needs, such as The reconstruction descriptor of the corresponding position and the reconstruction descriptor of the opposite feature.

Step S102, calculating residual coefficients of the prediction signal.

Preferably, in this step, the prediction residual is calculated by means of difference.

Step S104, quantizing the residual coefficients to obtain quantized coefficients.

The quantization method in this step may be scalar quantization or vector quantization.

Step S105, performing entropy coding on the quantized coefficients, and outputting a code stream.

For entropy coding in this step, preferably, the coding model adopts a context-adaptive binary entropy coding method (CABAC, Context-Adaptive Binary Arithmetic Coding) commonly used in video coding standards, and finally forms a binary code stream.

Through the video local feature descriptor compression method described in this embodiment, the video local feature descriptor is compressed, and on the basis of joint video content, the compressed video data is expressed in a compact manner, achieving a very high compression ratio , thereby improving the transmission rate and storage efficiency of video data, and at the same time improving the retrieval efficiency of video data.

Fig. 2 is a schematic flowchart of a method for compressing video local feature descriptors according to the second embodiment of the present invention.

As shown in Figure 2, steps S201, S202, and S205 in this embodiment are basically the same as steps S102, S102, S104, and S105 in the first embodiment. The difference is that the method for compressing local feature descriptors in this embodiment Among them, step S203 is also included after step S202 and before step S204, and the corresponding step S204 is different at the same time, specifically:

Step S203, transforming the residual coefficients to obtain transformation coefficients.

Preferably, the transformation method in this step is DCT transformation, KLT transformation or DST transformation transformation, so as to obtain transformation coefficients. The transformation here can use a one-dimensional coefficient matrix or a two-dimensional coefficient matrix.

Step S204, quantize the transform coefficients to obtain quantized coefficients. The quantization here can be scalar quantization or vector quantization.

Fig. 3 is a flowchart of a compression method in a specific example of the second embodiment of the present invention.

This embodiment takes a 128-dimensional SIFT local feature descriptor of video data as an example to describe the compression method of the present invention in detail.

As shown in Figure 3, the video-based local feature descriptor compression method of the present embodiment includes the following steps:

In step S301, firstly, the video content carried by the original video frame is encoded by a video encoder to obtain motion vector information of the video.

Step S302, extracting local feature descriptors from the original video frame, specifically:

Assume that the j-th local feature descriptor of the i-th frame currently being encoded is in is the spatial location information of the descriptor, is the local feature descriptor vector.

Step S303, performing inter-frame prediction on the extracted local feature descriptors, by formula (1)

$<span\; class="notranslate">\; (</span>{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; v</span>}}}_{\mathrm{<span\; class="notranslate">\; inter</span>}}<span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; v</span>}}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Psi;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; \&Psi;</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; min</span>}}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; v</span>}}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 1</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>$

Perform inter-frame prediction, namely:

First get the current position in the video encoder The motion vector MV(d _i , d _x , d _y ), d _i , d _x , d _y represent the offset value (di) of the time dimension and the offset value (dx,dy) of the two-dimensional space coordinates, respectively. Set in (id _i ) ^th frame with In the search range Ψ (including K _Ψ feature points) where the coordinates are the starting point, the inter-frame prediction signal is obtained, that is, the prediction signal

Step S304, making a difference between the original signal and the predicted signal to obtain the residual coefficient

Step S305, for the residual coefficient Carry out the transformation as steps S305A1～S305C1:

In step S305A1, the 128-dimensional Divided into two 8x8 two-dimensional matrices;

Step S305B1, using two 8x8 DCT transformations to act on the two transformation matrices respectively, to obtain two transformed coefficient matrices;

Step S305C1, combining two 8x8 coefficient matrices into one 128-dimensional transformation coefficient

Alternatively, the residual coefficients are transformed as in steps S305A2 to S305C2:

In step S305A2, the 128-dimensional According to 8 different directions, it is divided into 8 4x4 two-dimensional matrices, that is, the 16 elements in each direction are arranged into a 4x4 matrix according to the position;

Step S305B2, using 8 4x4 DCT transforms to act on the 8 transform matrices respectively, to obtain 8 transformed coefficient matrices;

Step S305C2, combining eight 4x4 coefficient matrices into one 128-dimensional transformation coefficient

Step S305 may also include:

Step S305D, decide whether to perform transformation through rate-distortion selection, if the transformation is skipped, then the residual coefficient is the transformation coefficient to be quantized at this time. When the transformation is skipped and the residual coefficients are directly quantized, it is a specific example of the first embodiment.

Step S306, perform scalar quantization on the transform coefficients by the set quantization step size (Qs, Quantization step) Get the quantization coefficient after quantization The quantization here may also be vector quantization, and scalar quantization is preferred in this embodiment.

Step S307, quantize the coefficients by an entropy encoder Entropy coding is carried out, and the coding model adopts the context-adaptive binary entropy coding method (CABAC, Context-Adaptive Binary Arithmetic Coding) commonly used in video coding standards, and finally forms a binary code stream, and the binarization process in entropy coding Specific steps include:

Step S307A, output a bin representation Whether it is an all-zero coefficient, if it is the end of the algorithm;

Step S307B, output a binary vector, wherein each bin represents Whether the corresponding coefficient in is zero;

Step S307C, for each non-zero coefficient, first output a sign bit, then output a bin to indicate whether the absolute value of this coefficient is 1; if not output a bin to indicate whether the absolute value of the coefficient is 2; if not, set this Exponential Golomb code output after subtracting 3 from the absolute value of the coefficient. After the encoding is completed, the reconstructed descriptor is obtained, and the code stream is output.

Step S308, for quantized quantized coefficients Perform inverse quantization and/or inverse transformation to obtain a reconstructed descriptor and store it in the cache.

It should be noted that when entropy encoding is performed in step S307, entropy encoding is also performed on the transformation control flag at the same time, and the transformation process done is recorded in the encoding. The transformation control flag here refers to whether the transformation in step S305 and the inverse transformation in step S308 have been performed. The transformation and inverse transformation here appear accordingly. If the transformation is performed, when the descriptor is reconstructed, the inverse transformation needs to be performed after dequantization; if no transformation is performed during the compression process, when the descriptor is reconstructed, There is no need for inverse transform after inverse quantization.

The video-based local feature descriptor compression method described in this embodiment compresses the video local feature descriptor, and on the basis of joint video content, the compressed video data can be represented in a compact manner, achieving a very high Compression ratio, thereby improving the transmission rate and storage efficiency of video data, and at the same time improving the retrieval efficiency of video data.

Fig. 4 is a schematic structural diagram of a video-based local feature descriptor compression system according to the third embodiment of the present invention.

As shown in FIG. 4 , the compression system based on video local feature descriptors in this embodiment includes: a prediction module 11, a residual coefficient calculation module 12, a quantization module 14, and an encoding module 15; wherein,

The prediction module 11 is used to perform inter-frame prediction on the current local feature descriptor in conjunction with video content to obtain a prediction signal.

The prediction module 11 performs inter-frame prediction on the current local feature descriptor, specifically, the prediction module 11 finds a reconstructed descriptor closest to the current local feature descriptor in the encoded frame of the video , as a predictive signal. Video data is composed of frames, and each frame has a corresponding local feature descriptor. The frame to which the current local feature descriptor in the process of compression belongs, that is, the current frame. Preferably, the encoded frame of the video is the previous frame of the current frame. The closest here means that the encoded reconstructed descriptor has the most identical feature data with the current descriptor, and it can also be selected through the sum of absolute value errors (SAD, Sum of Absolute Difference). This optimal value selection can be realized by a programmable logic controller (PLC). When the prediction module 11 searches for the closest reconstructed descriptor in the previous frame, it can search in all the data of the previous frame, or can customize the search range, such as searching according to the motion vector of the video, in the current description At the position pointed by the sub-motion vector, the closest reconstruction descriptor is searched within the range of the five nearest nodes.

In particular, when the prediction module 11 performs inter-frame prediction, the reconstructed descriptor as the prediction result of the current local feature descriptor, in addition to the closest reconstructed descriptor within the corresponding range, can also perform other reconstructions as required. Definition, such as the reconstruction descriptor of the corresponding position and the reconstruction descriptor of the opposite feature.

The residual coefficient calculation module 12 is connected to the prediction module 11 for calculating the residual coefficient of the prediction signal. Preferably, in this step, the residual coefficient of the predicted signal is calculated by means of difference.

The quantization module 14 is connected to the prediction selection module 12, and is used to receive the residual coefficient output by the prediction module 12, and is used to quantize the residual coefficient to obtain and output a quantized coefficient. The quantization module 14 performs scalar quantization or vector quantization on the residual coefficients.

The encoding module 15 is connected to the quantization module 14, and is used for receiving the quantization coefficient output by the quantization module 14, and for performing entropy encoding on the quantization coefficient, and outputting a code stream.

Through the video local feature descriptor compression system described in this embodiment, the video local feature descriptor is compressed, and on the basis of joint video content, the compressed video data is expressed in a compact manner, achieving a very high compression ratio , thereby improving the transmission rate and storage efficiency of video data, and at the same time improving the retrieval efficiency of video data.

FIG. 5 is a schematic structural diagram of a video-based local feature descriptor compression system according to a fourth embodiment of the present invention.

As shown in FIG. 5 , the descriptor compression system in this embodiment and the local feature descriptor compression system in the third embodiment include the same prediction module 11, residual coefficient calculation module 12, and encoding module 15. Notably, this embodiment also includes a transform module 13 and a reconstructed descriptor storage module 16, while the connection relationship of the quantization module 14 is different, specifically:

The transformation module 13 is connected to the residual coefficient calculation module 12 and the quantization module 14, and is used to receive the residual coefficient output by the residual coefficient calculation module 12, and is used to transform the residual coefficient to obtain and output transform factor.

The quantization module 14 is used for receiving the transform coefficients output by the transform module 13, and for quantizing the transform coefficients to obtain quantized coefficients.

The reconstructed descriptor storage module 16 is connected to the encoding module 15 and the predicting module 11, and is used for performing inverse quantization and inverse transformation on the quantized coefficients to obtain a reconstructed descriptor and store the reconstructed descriptor.

FIG. 6 is a schematic flowchart of a video compression method according to a fifth embodiment of the present invention.

As shown in Figure 6, the video compression method described in this embodiment includes:

Step S400, compressing the original video data to obtain the code stream of the video itself, and obtain the motion vector information of the video.

Step S401 , inter-frame prediction is performed on the current local feature descriptor in conjunction with video content to obtain a prediction signal.

In this step, inter-frame prediction is performed on the current local feature descriptor, specifically, in the coded frame of the video, according to the motion vector information, find a descriptor that is closest to the current local feature descriptor The reconstructed descriptor of is used as a prediction signal.

Step S402, calculating residual coefficients of the prediction signal. Preferably, in this step, the residual coefficient of the predicted signal is calculated by means of difference.

Step S403, transforming the residual coefficients to obtain transformation coefficients.

In this step, preferably, the transformation method is to obtain a coefficient matrix by applying DCT to a two-dimensional matrix, thereby obtaining transformation coefficients.

Step S404: Quantize the transform coefficient after transforming the residual coefficient to obtain a quantized coefficient. Preferably, the quantization here is scalar quantization.

Step S405, performing entropy coding on the quantized coefficients, completing the compression of local feature descriptors in the original video data, and obtaining entropy coded descriptor streams. For the entropy coding in this step, preferably, the coding model adopts a context-adaptive binary entropy coding method (CABAC, Context-Adaptive Binary Arithmetic Coding) commonly used in video coding standards, and finally forms a binary descriptor stream.

Step S406, output the description sub-code stream into the video code stream itself, and output the video code stream.

Through the video compression method described in this embodiment, the video is compressed, and the local feature descriptor of the video is compressed on the basis of joint video content, so that the compressed video data can be represented in a compact manner, and a high compression ratio can be achieved, thereby The transmission rate and storage efficiency of the video data are improved, and the retrieval efficiency of the video data is also improved.

Those of ordinary skill in the art can understand that all or part of the steps and modules for realizing the method and system of the present invention can be completed by hardware, or by program instruction hardware, and the program can be stored in a computer-readable storage medium, such as a disk , storage, etc.

It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principle of the present invention, and not to limit the present invention. Therefore, any modification, equivalent replacement, improvement, etc. made without departing from the spirit and scope of the present invention shall fall within the protection scope of the present invention. Furthermore, it is intended that the appended claims of the present invention embrace all changes and modifications that come within the scope and metesques of the appended claims, or equivalents of such scope and metes and bounds.

Claims (10)

1. a compression method of video local feature descriptor, is characterized in that, described method comprises the steps: Combine video content to perform inter-frame prediction on the current local feature descriptor to obtain a prediction signal; calculating residual coefficients of said predicted signal; Quantizing the residual coefficients to obtain quantized coefficients; Entropy encoding is performed on the quantized coefficients to output a code stream. 2. the compression method of video local feature descriptor according to claim 1, is characterized in that, after described calculating residual coefficient, before obtaining quantization coefficient, described method also comprises: transforming the residual coefficients to obtain transformation coefficients; Quantizing the residual coefficients to obtain quantized coefficients, and further, quantizing the transformation coefficients to obtain quantized coefficients. 3. The method for compressing video local feature descriptors according to claim 1, wherein the method further comprises: performing inverse quantization and inverse transformation on the quantization coefficients to obtain reconstruction descriptors, and storing the Refactoring descriptors. 4. The method for compressing video local feature descriptors according to claim 3, wherein the joint video content performs inter-frame prediction on the current local feature descriptor to obtain a prediction signal, further comprising: In the coded previous frame of the frame where the descriptor is located, a reconstructed descriptor closest to the current local feature descriptor is found as a prediction signal. 5. The method for compressing video local feature descriptors according to claim 4, wherein said quantization is scalar quantization or vector quantization. 6. A compression system for video local feature descriptors, characterized in that the system includes: a prediction module, a residual coefficient calculation module, a quantization module, and an encoding module The prediction module is used to perform inter-frame prediction on the current local feature descriptor in conjunction with video content to obtain a prediction signal; The residual coefficient calculation module is connected to the prediction module, and is used to calculate the residual coefficient of the prediction signal; The quantization module is connected to the residual coefficient calculation module, and is used to quantize the residual coefficient to obtain a quantized coefficient; The coding module is connected to the quantization module, and is used for performing entropy coding on the quantization coefficients and outputting a code stream. 7. the compression system of video local feature descriptor according to claim 6, is characterized in that, described system also comprises transformation module, and described transformation module is connected with described residual coefficient calculation module and quantization module, is used for Transforming the residual coefficients to obtain transformation coefficients; The quantization module is also used to quantize the transform coefficients and obtain quantized coefficients. 8. the compression system of video local feature descriptor according to claim 6, is characterized in that, described system also comprises: Reconstruction descriptor storage module, is used for dequantization and inverse transformation are carried out to described quantization coefficient, obtains reconstruct the descriptor, and store the reconstructed descriptor. 9. The compression system of the video local feature descriptor according to claim 8, wherein the prediction module is further used to find a coded previous frame of the frame where the current local feature descriptor is located. The reconstructed descriptor closest to the current local feature descriptor is used as a prediction signal. 10. A video compression method, characterized in that the method comprises: Compress the original data of the video to obtain the code stream of the video itself; In combination with the content of the video, the local feature descriptor in the original video data is compressed by using the compression method of the video local feature descriptor described in any one of claims 1 to 5 to obtain an entropy-coded descriptor stream ; Output the description sub-code stream into the video code stream itself, and output the video code stream.