CN111274950A - Feature vector data encoding and decoding method, server and terminal - Google Patents

The application discloses a feature vector data coding and decoding method, a server and a terminal. In the application, a server obtains an image feature vector in an image database, wherein the image feature vector is floating point type data; the server converting the floating-point data into a binary data text comprising a binary encoding of ASCII characters, wherein a valid bit in a floating-point data corresponds to a binary encoding of an ASCII character; the server encodes the binary code of each ASCII character in the binary data text to obtain an encoded binary data text; the server compresses the encoded binary data text.

Feature vector data encoding and decoding method, server and terminal

Technical Field

The present application relates to the field of communications, and in particular, to a feature vector data encoding and decoding method, a server, and a terminal.

Background

With the wide application of image recognition technology, the extraction amount of image feature data is increasing. Taking face recognition based on a deep learning algorithm as an example, when extracting features of an image, feature vector data to be extracted is generally 128, 512 or 1024 floating point (float) data.

One floating point type data (including single-precision floating point number or double-precision floating point number) occupies 4 bytes, and for example, 512 feature vectors are extracted from a single image, and the size of the feature vector data extracted from the single image reaches 2048 bytes. Taking the face database capable of storing 30 ten thousand face pictures as an example, the storage capacity of the feature vector data can reach 60MB, and for a larger face database, such as a face database with the scale from million figures to hundred million figures, the storage capacity occupied by the feature vector data is larger, and the network transmission overhead is higher.

Therefore, how to improve the coding efficiency of the feature vector data, thereby reducing the storage amount of the feature vector data and further reducing the overhead of network transmission is a problem to be solved at present.

Disclosure of Invention

The exemplary embodiments of the present application provide a feature vector data encoding and decoding method, a server, and a terminal device, so as to improve the encoding efficiency of feature vector data.

According to an aspect of the exemplary embodiments, there is provided a feature vector data encoding method including:

the method comprises the steps that a server obtains image feature vectors in an image database, wherein the image feature vectors are floating point data;

the server converting the floating-point data into a binary data text comprising a binary encoding of ASCII characters, wherein a valid bit in a floating-point data corresponds to a binary encoding of an ASCII character;

the server encodes the binary code of each ASCII character in the binary data text to obtain an encoded binary data text;

the server compresses the encoded binary data text.

In some exemplary embodiments, the server converts the floating point type data into a binary data text, including: removing invalid bits in the floating-point data by the server to obtain valid bits in the floating-point data, wherein the invalid bits in the floating-point data comprise all zeros after the last non-zero bit after a decimal point; the server converts each valid bit into a binary encoding of the corresponding ASCII character according to an ASCII character table.

In some exemplary embodiments, the server encodes a binary encoding of each ASCII character in the binary data text, including:

the server takes each ASCII character in the binary data text as a node, and sorts all the nodes according to the weight of each node, wherein the weight of one node is used for representing the probability of the ASCII character corresponding to the node appearing in all the ASCII characters in the binary data text;

and the server obtains the binary coding value corresponding to each leaf node according to the coding value corresponding to the node on the path of each leaf node.

In some exemplary embodiments, the method further comprises: the server sends the compressed binary data text and the coding table to a terminal, so that the terminal decodes the binary data text according to the coding table to obtain a feature vector of an image; wherein the encoding table includes a mapping between binary encodings of the ASCII characters and encoded values resulting from binary encodings of the corresponding ASCII characters.

According to an aspect of the exemplary embodiments, there is provided a feature vector data decoding method including:

a terminal receives feature vector data and a coding table sent by a server, wherein the feature vector data is a compressed and coded binary data text; wherein the encoding table includes a mapping relationship between binary encodings of the ASCII characters and encoded values resulting from binary encodings of the corresponding ASCII characters;

the terminal decompresses the binary text data, and the decompressed binary text comprises a coded value obtained by coding the binary code of the ASCII characters;

the terminal decodes the decompressed binary text data according to the coding table to obtain a binary code of each ASCII character in the binary text;

and the terminal converts the decoded binary text data into floating-point type data to obtain a floating-point type image characteristic vector, wherein one effective bit in one piece of floating-point type data corresponds to a binary code of an ASCII character.

According to an aspect of the exemplary embodiments, there is provided a server including:

the acquisition module is used for acquiring image feature vectors in an image database, wherein the image feature vectors are floating point data;

a conversion module for converting the floating-point data into a binary data text comprising a binary encoding of ASCII characters, wherein a significant bit in a floating-point data corresponds to a binary encoding of an ASCII character;

the encoding module is used for encoding the binary code of each ASCII character in the binary data text to obtain an encoded binary data text;

and the compression module is used for compressing the coded binary data text.

In some exemplary embodiments, the conversion module is specifically configured to: removing invalid bits in the floating-point data to obtain valid bits in the floating-point data, wherein the invalid bits in the floating-point data comprise all zeros after the last non-zero bit after a decimal point; each valid bit is converted to a binary encoding of the corresponding ASCII character according to the ASCII character table.

In some exemplary embodiments, further comprising: the transmitting module is used for transmitting the compressed binary data text and the coding table to a terminal, so that the terminal decodes the binary data text according to the coding table to obtain a feature vector of an image; wherein the encoding table includes a mapping between binary encodings of the ASCII characters and encoded values resulting from binary encodings of the corresponding ASCII characters.

According to an aspect of the exemplary embodiments, there is provided a terminal including:

the receiving module is used for receiving the feature vector data and the coding table sent by the server, wherein the feature vector data is a compressed and coded binary data text; wherein the encoding table includes a mapping relationship between binary encodings of the ASCII characters and encoded values resulting from binary encodings of the corresponding ASCII characters;

the decompression module is used for decompressing the binary text data, and the decompressed binary text comprises a coded value obtained by coding the binary code of the ASCII characters;

the decoding module is used for decoding the decompressed binary text data according to the coding table to obtain a binary code of each ASCII character in the binary text;

and the conversion module is used for converting the decoded binary text data into floating-point type data to obtain a floating-point type image characteristic vector, wherein one effective bit in one piece of floating-point type data corresponds to the binary code of one ASCII character.

According to an aspect of the exemplary embodiments, there is provided a server including: a processor and a memory, the memory storing computer instructions, the processor being configured to execute the computer instructions to implement the method performed by the server described above.

According to an aspect of the exemplary embodiments, there is provided a terminal including: a processor and a memory, the memory storing computer instructions, the processor being configured to execute the computer instructions to implement the method performed by the terminal described above.

According to an aspect of the exemplary embodiments, there is provided a computer-readable storage medium having stored thereon computer instructions to cause a processor to execute the computer instructions to implement the above-described server-side method.

According to an aspect of exemplary embodiments, there is provided a computer-readable storage medium having stored thereon computer instructions to cause a processor to execute the computer instructions to implement the above-described terminal-side method.

In the above embodiment of the present application, the floating-point type data is converted into a binary data text comprising a binary code of ASCII characters, wherein one significant bit in one floating-point type data corresponds to the binary code of one ASCII character; and then, coding the binary code of each ASCII character in the binary data text to obtain a coded binary data text, and then compressing the coded binary data text, thereby improving the coding efficiency of the feature vector data, reducing the storage capacity of the feature vector data and further reducing the network transmission overhead of the feature vector data.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to these drawings without inventive exercise.

Fig. 1 schematically illustrates a flow chart of an encoding method of feature vector data provided by an embodiment of the present application;

fig. 2a and 2b are schematic diagrams illustrating huffman coding in the embodiment of the present application;

fig. 3 is a schematic flowchart illustrating a decoding method of feature vector data according to an embodiment of the present application;

fig. 4 is a schematic structural diagram illustrating a server provided in an embodiment of the present application;

fig. 5 schematically illustrates a structure of a terminal provided in an embodiment of the present application.

Detailed Description

The technical solution in the embodiments of the present application will be described in detail and removed with reference to the accompanying drawings. Wherein in the description of the embodiments of the present application, "/" means or, unless otherwise stated, for example, a/B may mean a or B; "and/or" in the text is only an association relationship describing an associated object, and means that three relationships may exist, for example, a and/or B may mean: three cases of a alone, a and B both, and B alone exist, and in addition, "a plurality" means two or more than two in the description of the embodiments of the present application.

The terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as implying or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of embodiments of the application, "plurality" means two or more unless stated otherwise.

Furthermore, the terms "comprises" and "comprising," as well as any variations thereof, are intended to cover a non-exclusive inclusion, such that a product or device that comprises a list of elements is not necessarily limited to those elements explicitly listed, but may include other elements not expressly listed or inherent to such product or device.

The term "module" as used herein refers to any known or later developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and/or software code that is capable of performing the functionality associated with that element.

Fig. 1 schematically illustrates a feature vector encoding process provided by an embodiment of the present application, and as shown in the figure, the process may include:

s101: the server obtains image feature vectors in an image database, wherein the image feature vectors are floating point data.

In this step, the server may obtain the image feature vector in the image database and perform subsequent encoding processing when the image feature vector in the image database needs to be sent to the terminal. The server may also obtain the image feature vector in the updated image database after the image database is updated, and perform subsequent encoding processing. The server can also obtain the image feature vector in the image database based on the request of the terminal for obtaining the image feature vector, and perform subsequent coding processing.

The image database may be a face database for storing face images and face feature vectors extracted from the face images.

Wherein, the server can be a server with image database management function, and the server can also be responsible for the distribution of the image feature vector. The terminal can perform image recognition based on the image feature vector, for example, face recognition based on the face feature vector. In some application scenarios, the terminal may be a face acquisition and recognition device deployed at an entrance of a specific place (such as an airport or a station); in other application scenarios, the terminal may also be a mobile handheld device for face acquisition and recognition.

S102: the server converts the floating-point data to binary data text comprising a binary encoding of ASCII characters, wherein a valid bit in a floating-point data corresponds to a binary encoding of an ASCII character.

Optionally, in this step, the server may first remove an invalid bit in the floating-point data to obtain a valid bit in the floating-point data, where the invalid bit in the floating-point data includes all zeros after a last non-zero bit after the decimal point; the server then converts each valid bit to a binary encoding of the corresponding ASCII character according to the ASCII character table.

The decimal representation range for a floating point type number is: 0.000000000000000000000000000000000000034-340000000000000000000000000000000000000. The binary representation of a decimal number (e.g., 313645) is 1001100100100101101. In 0.31364500000000000000000000000000000000, in a floating-point type value, the 0 following the last non-zero bit "5" following the decimal point can be regarded as invalid bit and can be discarded without any loss of valid information, so that each bit (including the bit where the decimal point is located) in 0.313645 can be converted into corresponding ASCII character according to the ASCII character table, thereby obtaining 8 bytes of ASCII code data. Where the decimal point may be converted to a period in an ASCII character table.

S103: and the server encodes the binary code of each ASCII character in the binary data text to obtain the encoded binary data text.

In this step, a binary code of each ASCII character in the binary data text may be encoded using a huffman (huffman) encoding algorithm, or a gzip encoding algorithm, or a brotli encoding algorithm, or a zstd encoding algorithm, or a PPM encoding algorithm to compress the storage amount of the binary data text.

S104: the server compresses the encoded binary data text.

By further compression processing, the amount of memory of the text of the progress data can be further compressed. In this step, the encoded binary data text may be compressed losslessly according to the encoding table, and the range of the text compression may be 10 times to 1000 times, so as to offset the data size amplified when the floating point type data is converted into ASCII characters in S102.

Further, the above process may further include the following steps:

s105: and the server sends the compressed binary data text and the coding table to the terminal, so that the terminal decodes the binary data text according to the coding table to obtain the feature vector of the image. Wherein the encoding table includes a mapping between binary encodings of the ASCII characters and encoded values resulting from binary encodings of the corresponding ASCII characters. Taking huffman coding algorithm as an example, the coding table may be a huffman tree structure, where each leaf node corresponds to one ASCII character, and the corresponding ASCII character may be determined based on the tree according to the coding value.

Next, the encoding process in S103 in the above flow will be described by taking the huffman coding algorithm as an example.

The principle of huffman coding is explained in the following with an example.

In this example, the ASCII character sequence is: "AFTERDATAEARAREARTAREA", the ASCII character sequence may be encoded by:

firstly, counting the occurrence frequency of each character as the weight of a corresponding node (one character corresponds to one node):

A：8，D：1，E：4，F：1，R：5，T：3。

then, a Huffman tree was constructed in the following way:

repeating the following steps:

sequencing all the nodes according to the weight to obtain: D-F-T-E-R-A;

selecting two nodes with the minimum weight, namely D and F, to generate a new node, wherein the weight of the node is the sum of the weights of the two nodes, namely 2;

and then, returning to the step of sequencing all the nodes according to the weight until only one node is left, wherein the node is the root node of the Huffman tree.

According to the above steps, the generation process of the Huffman tree corresponding to the ASCII character sequence can be seen in fig. 2 a.

According to the Huffman tree generated in fig. 2a, each leaf node in the number corresponds to an ASCII character in the sequence of ASCII characters.

Based on the above-described Huffman tree structure, each ASCII character may be encoded using the Huffman tree. Specifically, the following can be agreed: and taking the node with small weight as a left node, taking the node with large weight as a right node, wherein the left node is coded into 0, and the right node is coded into 1. Thus, the above-described encoding format is shown in fig. 2 b. As can be derived from the graph of fig. 2b, the code value of each leaf node is:

E：00；R：01；F：1000；D：1001；T：101；A：11。

in the above example, the ASCII character sequence including letters is taken as an example, and after the image feature vector of the floating point type is converted into the ASCII character sequence, the ASCII character sequence is a sequence including decimal digits, but the encoding principle and method thereof are the same as those described above.

Fig. 3 schematically illustrates a flow chart of feature vector data decoding provided by an embodiment of the present application, and as shown in the figure, the flow chart may include the following steps:

s301: a terminal receives feature vector data and a coding table sent by a server, wherein the feature vector data is a compressed and coded binary data text; wherein the encoding table includes a mapping between binary encodings of the ASCII characters and encoded values resulting from binary encodings of the corresponding ASCII characters.

S302: the terminal decompresses the binary text data, the decompressed binary text comprising encoded values encoding the binary encoding of the ASCII characters.

S303: and the terminal decodes the decompressed binary text data according to the coding table to obtain the binary code of each ASCII character in the binary text.

S304: and the terminal converts the decoded binary text data into floating-point type data to obtain a floating-point type image characteristic vector, wherein one effective bit in one floating-point type data corresponds to a binary code of an ASCII character.

The decoding process at the terminal side corresponds to the encoding process at the server side, and the related description can refer to the encoding process at the server side.

In the above embodiment of the present application, the floating-point type data is converted into a binary data text comprising a binary code of ASCII characters, wherein one significant bit in one floating-point type data corresponds to the binary code of one ASCII character; and then, coding the binary code of each ASCII character in the binary data text to obtain a coded binary data text, and then compressing the coded binary data text, thereby improving the coding efficiency of the feature vector data, reducing the storage capacity of the feature vector data and further reducing the network transmission overhead of the feature vector data.

Based on the same technical concept, the embodiment of the application also provides a server.

Based on the same technical concept, the embodiment of the application also provides a terminal.

Since the communication terminal and the computer storage medium in the embodiment of the present application may be applied to the processing method, reference may also be made to the above method embodiment for obtaining technical effects, and details of the embodiment of the present application are not described herein again.

Those of ordinary skill in the art will understand that: all or a portion of the steps of implementing the above-described method embodiments may be performed by hardware associated with program instructions. The program may be stored in a computer-readable storage medium. When executed, the program performs steps comprising the method embodiments described above; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.

While specific embodiments of the present application have been described above, it will be appreciated by those skilled in the art that these are by way of example only, and that the scope of the present application is defined by the appended claims. Various changes and modifications to these embodiments may be made by those skilled in the art without departing from the spirit and principles of this application, and these changes and modifications are intended to be included within the scope of this application.