JP6104209B2 - Hash function generation method, hash value generation method, apparatus, and program - Google Patents

Description

  The present invention relates to a hash function generation method, a hash value generation method, an apparatus, and a program, and more particularly, to a hash function generation method, a hash value generation method, an apparatus, and a program for generating a hash function.

  The number of contents (images, videos, sounds, documents, etc.) distributed on the network has become enormous due to the advancement and high quality of communication environments, computers, and distributed processing infrastructure technologies. For example, a search engine is said to have trillions of web pages indexed. Some sites report that 350 million images are uploaded every day, and some sites report that 64 hours of video per minute are newly released. .

  Such an enormous amount of content is a rich source of information for users, but also brings about the problem that it becomes increasingly difficult to quickly access the content to be viewed. In such a trend, there is an increasing demand for media analysis technology for efficiently searching for contents to be browsed and viewed.

  In content analysis, the discovery of similar content plays an important role. For example, when classifying content, similar content is classified into the same category. Alternatively, in the case of search, when a certain content is given as a query, it is a basic requirement to search for content similar to this content. In addition, in content recommendation, it is redundant to find and recommend content similar to the content that the user has browsed / viewed so far, and even in the case of content summary, it is redundant to present similar content Therefore, it is necessary to perform processing to find and omit this.

  Here, a typical procedure for finding similar contents will be described. First, the content is expressed by a certain feature amount. Next, the similarity is calculated by measuring the proximity of the feature quantities, and similar content is found based on the similarity. To give a simple example, if the content is an image or video, the degree of similarity can be measured using the color histogram of the image (video frame) as a feature amount. If the content is a document, the degree of similarity can be measured using a histogram of the appearance frequency of words (referred to as a Bag-of-Words histogram) as a feature amount. Needless to say, if the number of contents is 1000, it is necessary to calculate the similarity for each of the 1000 contents, and to pick up the content having a high similarity as a similar content.

  However, as described above, when a huge amount of content is targeted, there is a problem of consuming a large amount of calculation time and memory.

In general, the dimension of the feature amount (vector) of the content is often high, and the calculation of the similarity requires enormous time. In general, the dimension of the Bag-of-Words histogram of a document is the same as that of a word type (vocabulary). Even a simple feature quantity such as an image color histogram is generally a real-valued vector of hundreds to thousands of dimensions, and several hundreds of thousands to several hundreds of recently used feature expressions based on sparse and Fisher kernels. It can also be a million-dimensional vector. Furthermore, since it is necessary to calculate the degree of similarity for all content sets, no matter what degree of similarity calculation means is used, assuming that there are D feature quantities and N contents, O (DN ). In order to execute an immediate search, it is preferable to store a feature amount or its similarity in a memory. However, in order to perform this, an O (N ² ) memory is required. In this way, it is necessary to deal with content on the order of more than 100 million, and unrealistic time and memory are required.

  In order to solve such a problem, several inventions have been conventionally made.

  For example, in the technology disclosed in Non-Patent Document 1, a hash function group is generated such that the similarity between the original feature amount and the collision probability are equal in any two adjacent contents (feature amounts). A cosine similarity is considered as a typical similarity, and the basic procedure for generating a hash function in that case is by generating a plurality of random hyperplanes in the feature amount space (called random projection). By hashing the feature amount depending on which side of each hyperplane the feature amount exists, similar content can be found approximately without obtaining similarity between all the contents.

  Further, the technique disclosed in Non-Patent Document 2 captures a distribution of feature amounts and configures an optimum hash value for the distribution. Specifically, by capturing the manifold structure in the feature space and finding non-linear embedding in the binary space (hash space) that optimally stores the manifold structure, the original high-dimensional feature is reduced. Convert to a bit hash value. By evaluating the similarity of the hash values, high-speed similar content can be found.

  Further, in the technique disclosed in Patent Document 1, a hash function is obtained based on content information and related information (correct data) indicating whether or not two different contents should be associated with each other. The feature value is converted into a low-bit hash value.

JP 2013-68884 A

M. Datar, N. Immorlica, P. Indyk, V.S. Mirrokni, “Locality-Sensitive Hashing Scheme based on p-Stable Distributions”, In Proceedings of the Twentieth Annual Symposium on Computational Geometry, 2004, p.253-262. Go Irie, Zhengu Li, Shih-Fu Chang, “Hashing to Preserve Structure”, Image Recognition and Understanding Symposium, 2013.

  In the technologies disclosed in Non-Patent Documents 1 and 2, the original content is converted into a compact hash value, thereby making it possible to find similar content with very high accuracy and high speed. However, in any of the techniques, the hash value is obtained only from the content feature amount information, and the hash value is not generated in accordance with the semantic viewpoint of the content, so that sufficient accuracy cannot be obtained. There was a problem.

  On the other hand, in the technology disclosed in Patent Document 1, a hash function is obtained based on related information (correct data) indicating whether or not two different contents should be associated with the feature amount of the content. It is possible to generate a hash value in accordance with.

  However, the related information as to whether or not the content should be associated with the semantic viewpoint of the content does not necessarily represent the same information, and the latter includes more various and various types. As an easy-to-understand example, it is assumed that the image has a total of four images, for example, “red apple”, “green apple”, “mandarin orange”, and “tuna”. In the technique of Patent Document 1, which image is related to which image is related (and is not) is given as related information. When these four images are considered, the same fruit “red apple” It is assumed that “green apple” is related, and other information is given that it is not related otherwise. Or “red apples”, “green apples” and “mandarin oranges” are fruits, so they are related, and it is considered that only “tuna” is not related and given relevant information. It could be taken. On the other hand, “red apples” and “tuna” can be considered to be weakly related from the viewpoint that both are special products of Aomori Prefecture. In other words, looking at “red apples” from various semantic viewpoints, they are strongly related to “green apples” in terms of fruits, weakly related to “mandarin oranges”, and weakly related to Aomori specialties. With the technology of Patent Document 1, it is impossible to express such various viewpoints and related strengths.

  However, until now, no technology can realize high-precision similar content discovery by constructing hash values based on various semantic viewpoints, while being fast and memory-saving. There was a problem.

  The present invention has been made in view of the above circumstances, and provides a hash function generation method, a hash value generation method, an apparatus, and a program that can generate a hash function for obtaining a hash value in consideration of the meaning of content. The purpose is to provide.

  In order to achieve the above object, a hash function generation method according to the present invention is a hash function generation method in a hash function generation apparatus including a feature extraction unit and a hash function generation unit, wherein the feature extraction unit includes a plurality of feature extraction units. For each of the contents, the step of extracting a feature amount from the content, and the hash function generation means for the feature amount of each of the plurality of contents extracted by the feature extraction means, and for each of the plurality of contents Based on a given semantic vector representing the meaning of the content, for each of the plurality of content, a hash value corresponding to a feature amount extracted from the content, and a hash value corresponding to the semantic vector of the content, So that the hash value corresponding to the feature amount is obtained. It is configured to include the steps of: generating a hash function for obtaining a hash value corresponding to the hash function, and the semantic vector of order, the.

  The hash function generation device according to the present invention includes, for each of a plurality of contents, a feature extraction unit that extracts a feature amount from the content, and the feature amount of each of the plurality of contents extracted by the feature extraction unit, Based on a semantic vector representing the meaning of the content given for each of the plurality of contents, a hash value corresponding to a feature amount extracted from the content for each of the plurality of contents, and the meaning of the content A hash function for generating a hash function for obtaining a hash value corresponding to the feature quantity and a hash function for obtaining a hash value corresponding to the semantic vector so that a distance from the hash value corresponding to the vector is reduced. Generating means.

  In the hash function generation method and the hash function generation device, the hash function generation unit is provided for the feature amount of each of the plurality of contents extracted by the feature extraction unit and each of the plurality of contents. Based on a semantic vector representing the meaning of the content, for each of the plurality of contents, in the feature amount space that is a space in which the feature amount exists, the feature amount of the content is set to the feature amount of the content. A hash value corresponding to the feature amount of the content obtained based on a manifold structure represented by a linear combination of hash values corresponding to the feature amount of other content existing in the vicinity, and a feature extracted from the content The distance from the hash value corresponding to the amount is reduced, and the plurality of contents For each, a hash for obtaining a hash value corresponding to the feature amount so that a distance between the hash value corresponding to the feature amount extracted from the content and the hash value corresponding to the semantic vector of the content becomes small A hash function for obtaining a function and a hash value corresponding to the semantic vector can be generated.

  In the hash function generation method and the hash function generation device, the hash function generation unit provides the feature amount of each of the plurality of contents extracted by the feature extraction unit and each of the plurality of contents. Based on the semantic vector representing the meaning of the content, the semantic vector of the content is the meaning of the content in a semantic vector space in which the semantic vector exists for each of the plurality of content. A hash value corresponding to the semantic vector of the content obtained based on a manifold structure represented by a linear combination of hash values corresponding to the semantic vector of other content existing in the vicinity of the vector, and a semantic vector of the content The distance from the hash value corresponding to is small And for each of the plurality of contents, the feature amount is reduced such that a distance between a hash value corresponding to the feature amount extracted from the content and a hash value corresponding to the semantic vector of the content is reduced. A hash function for obtaining a hash value corresponding to, and a hash function for obtaining a hash value corresponding to the semantic vector can be generated.

  The hash value in the present invention can be a binary value.

  A hash value generation method according to the present invention is a hash value generation method in a hash value generation device including a feature extraction unit and a hash value generation unit, wherein the feature extraction unit extracts a feature amount from content, The hash value generation unit is based on the feature amount extracted by the feature extraction unit and the hash function for obtaining a hash value corresponding to the feature amount generated by the hash function generation method of the present invention. And a step of generating a hash value corresponding to the feature amount of the content.

  The hash value generation device according to the present invention includes a feature extraction unit that extracts a feature amount from content, the feature amount extracted by the feature extraction unit, and the feature amount generated by the hash function generation device of the present invention. Hash value generation means for generating a hash value corresponding to the feature amount of the content based on the hash function for obtaining a corresponding hash value.

  The program of the present invention is a program for causing a computer to execute each step of the hash function generating method of the present invention or the hash value generating method of the present invention.

  As described above, according to the hash function generation method, apparatus, and program of the present invention, the feature amount of each of the plurality of contents and the semantic vector that represents the meaning of the content given to each of the plurality of contents. On the basis of each of the plurality of contents, the hash corresponding to the feature amount is set such that the distance between the hash value corresponding to the feature amount extracted from the content and the hash value corresponding to the semantic vector of the content is reduced. By generating a hash function for obtaining a value and a hash function for obtaining a hash value corresponding to a semantic vector, a hash function for obtaining a hash value in consideration of the meaning of the content can be generated. An effect is obtained.

  In addition, according to the hash value generation method, apparatus, and program of the present invention, the hash value corresponding to the feature amount of the content is generated based on the feature amount extracted from the content and the generated hash function. Thus, it is possible to obtain an effect that data similar to the content can be found with high accuracy by using the hash value in consideration of the meaning of the content.

It is the schematic which shows the structure of the information processing apparatus which concerns on the 1st Embodiment of this invention. It is explanatory drawing for demonstrating a hash function. It is explanatory drawing for demonstrating the manifold structure in feature-value space. It is explanatory drawing for demonstrating the hash function which considered the manifold structure in the feature-value space. It is explanatory drawing for demonstrating the manifold structure in a feature-value space, and the manifold structure in a semantic vector space. It is a flowchart which shows the content of the hash function generation process routine in the information processing apparatus which concerns on the 1st Embodiment of this invention. It is a flowchart which shows the content of the hash value generation process routine in the information processing apparatus which concerns on the 1st Embodiment of this invention. It is the schematic which shows the structure of the information processing system which concerns on the 3rd Embodiment of this invention. It is a figure which shows an example of the content matched by the hash value.

  In the embodiment of the present invention, a case where the present invention is applied to an information processing apparatus that generates a hash function and a hash value for realizing high-precision similar content discovery while being high-speed and memory-saving is taken as an example Explained. Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

<First Embodiment>
<System configuration>
In the embodiment of the present invention, a hash function and a hash value are generated for the content in consideration of a plurality of semantic viewpoints and the strength of the relation. The information processing apparatus 1 according to the first embodiment of the present invention generates a hash function, and generates a hash value using the generated hash function. The information processing apparatus 1 includes a computer including a CPU, a RAM, and a ROM that stores a program for executing a hash function generation processing routine and a hash value generation processing routine, which will be described later. As shown in FIG. As illustrated in FIG. 1, the information processing apparatus 1 includes an input unit 2, a calculation unit 3, and an output unit 4.

  A plurality of contents are registered in the contents database 5 shown in FIG. The content database 5 stores at least the content itself or an address that uniquely indicates the location of the content data. The content is, for example, a document file for a document, an image file for an image, a sound file for sound, a video file for video, and the like. It is assumed that an identifier capable of uniquely identifying the media type and itself is stored.

  Further, it is assumed that the semantic vector database 6 shown in FIG. 1 stores semantic vectors corresponding to the respective contents registered in the content database 5.

  The semantic vector represents the meaning of the content. The semantic vector is a vector given corresponding to each content, and expresses as a vector how much the content is related to one or more semantic concepts. For example, in the above example, the semantic vector can be defined as a three-dimensional vector representing three semantic concepts of “apple”, “fruit”, and “Aomori”.

"Red apple" is {apple: 1}, {fruit: 1}, {Aomori: 0.9}-> (1,1,0.9)

"Green apple" is {apple: 1}, {fruit: 1}, {Aomori: 0.1}-> (1,1,0.1)

"Mikan" is {apple: 0}, {fruit: 1}, {Aomori: 0}-> (0, 1, 0)

"Tuna" is {apple: 0}, {fruit: 0}, {Aomori: 0.7}-> (0, 0, 0.7)

And so on. According to the above three-dimensional semantic vector, the distance between “red apple” and “blue apple” is (1-1) ² + (1-1) ² + (0.9−0.1) ² = 0.64. Can be calculated. Similarly, the distance between “red apples” and “mandarin oranges” is 1.81, and the distance between “red apples” and “tuna” is 2.04. Can be expressed.

  Here, the most important point in the embodiment of the present invention is that a hash function and a hash value incorporating a semantic vector that can flexibly express a semantic viewpoint can be obtained. Depending on the technique of the above-mentioned Patent Document 1, only definite and uniform related information such as which content is related to which content can be considered, and thus a hash function and a hash value are obtained by incorporating a flexible semantic vector. Cannot be asked.

  If the semantic vector database 6 is not given, a separate semantic vector database input unit may be provided. The semantic vector database input unit is an input device for manually or mechanically providing a semantic concept corresponding to each dimension (element) of a semantic vector and a value of each element. When a general-purpose computer provided with a keyboard, a pointing device, or the like is mechanically given, it may be constituted by a general-purpose computer that can be connected to the Internet, for example.

  When giving manually, the user lists through the user interface the semantic concepts that are assumed to be closely related to the contents of the content, as shown in the previous example, and what semantic concept each content has. The degree to which the values are related may be given as a real value of 0 to 1, for example.

  Further, as an example of mechanically giving, for example, a case where an image is collected from a Web page can be cited. Most simply, for example, on the assumption that images and documents in the same Web page are related, a characteristic word included in the document is extracted and used as a semantic concept. Also good. Any known method may be used to extract characteristic words. However, after removing Stopwords (words with a low amount of information such as particles) on a rule basis, each called Document Frequency (DF) is used. You may perform by removing the word with the high / low number of documents containing a word.

  Alternatively, a classification keyword called a tag may be assigned to a specific Web page such as a blog, a news site, a microblog, or an SNS. In this case, these may be used directly as semantic concepts.

  In addition, metadata representing content details (content title, summary sentence, keyword), content format (content data amount, thumbnail size, etc.), and the like are stored in the semantic vector database 6. These may be used as semantic vectors.

  When the semantic vector is given mechanically, there is an advantage that the semantic vector can be obtained without manpower.

  The information processing apparatus 1 is connected to the content database 5 and the semantic vector database 6 via communication means, and communicates information with each other via the input unit 2 and the output unit 4. A hash function generation process for generating a hash function based on the registered content and a hash value generation process for converting the content into a plurality of binary values using the generated hash function are performed.

  In addition, the content database 5 and the semantic vector database 6 may be inside or outside the information processing apparatus 1, and any known communication means can be used. Suppose that the communication means is connected to communicate via the Internet or TCP / IP. The content database 5 and the semantic vector database 6 may be configured by so-called RDBMS (Relational Database Management System) or the like.

  Each unit of the information processing device 1, the content database 5, and the semantic vector database 6 may be configured by a computer, a server, or the like that includes an arithmetic processing device, a storage device, and the like, and the processing of each unit may be executed by a program. . This program is stored in a storage device included in the information processing apparatus 1, and can be recorded on a recording medium such as a magnetic disk, an optical disk, or a semiconductor memory, or provided through a network. Of course, any other component does not have to be realized by a single computer or server, and may be realized by being distributed to a plurality of computers connected by a network.

  Next, each part of the information processing apparatus 1 shown in FIG. 1 will be described.

  The input unit 2 acquires a plurality (N) of contents from the content database 5. Further, the input unit 2 accepts input of content as a search query. Further, the input unit 2 acquires a semantic vector given for each of the N contents from the semantic vector database 6.

  The calculation unit 3 includes a feature extraction unit 30, a hash function generation unit 32, a hash function storage unit 34, and a hash value generation unit 36.

The feature extraction unit 30 extracts a feature amount from each of the N pieces of content received by the input unit 2.
In addition, the feature extraction unit 30 extracts a feature amount from the content as a search query received by the input unit 2.

  The process of extracting feature amounts in the feature extraction unit 30 depends on the media type of the content. For example, the feature quantity that can be extracted or extracted varies depending on whether the content is a document, an image, a sound, or a video. Here, what kind of feature amount is extracted for each media type is not important as a requirement of the present embodiment, and a generally known feature extraction process may be used. Specifically, it is effective for all feature quantities as long as it is numerical data (scalar or vector) having a dimension extracted from a certain content. Accordingly, here, an example of feature extraction processing for various contents suitable for the present embodiment will be described.

  When the content is a document, the appearance frequency of words appearing in the document can be used. For example, the appearance frequency may be counted for each word corresponding to a noun, an adjective, or the like using a known morphological analysis. In this case, the feature amount of each document is expressed as a vector having the same dimensions as the word type.

  When the content is an image, for example, brightness features, color features, texture features, concept features, landscape features, shape features, and the like are extracted. The brightness feature can be extracted as a histogram by counting the V values in the HSV color space. In this case, the feature amount of each image is expressed as a vector having the same number of dimensions as the number of V-value quantizations (for example, 256 gradations for 16-bit quantization).

  The color feature can be extracted as a histogram by counting the values of the respective axes (L *, a *, b *) in the L * a * b * color space. The number of histogram bins on each axis may be, for example, 4 for L *, 14 for a *, 14 for b *, etc. In this case, the total number of bins for 3 axes is 4 × 14 × 14 = 784, that is, a 784-dimensional vector.

  As a texture feature, a statistic (contrast) of a density histogram, a power spectrum, or the like may be obtained. Alternatively, it is preferable to use a local feature amount because it can be extracted in the form of a histogram as in the case of color and movement. As the local feature, for example, SIFT (Scale Invariant Feature Transform) described in the following Reference 1 or SURF (Speeded Up Robust Features) described in the following Reference 2 can be used.

[Reference 1] DG Lowe, “Distinctive Image Features from Scale-Invariant Keypoints”, International Journal of Computer Vision, pp.91-110, 2004
[Reference 2] H. Bay, T. Tuytelaars, and LV Gool, “SURF: Speeded Up Robust Features”, Lecture Notes in Computer Science, vol. 3951, pp.404-417, 2006

  The local feature extracted by these becomes a 128-dimensional real value vector, for example. This vector is converted into a code with reference to a code length that has been learned and generated in advance, and a histogram can be generated by counting the number of the codes. In this case, the number of bins in the histogram matches the code number of the code length. Alternatively, the sparse expression described in Reference 3 or the feature expression based on the Fisher kernel described in References 4 and 5 may be used.

[Reference 3] Jinjun Wang, Jianchao Yang, Kai Yu, Fengjun Lv, Thomas Huang, and Yihong Gong, “Locality-constrained Linear Coding for Image Classification”, IEEE Conference on Computer Vision and Pattern Recognition, pp. 3360-3367, 2010.
[Reference 4] Florent Perronnin, Jorge Sanchez, Thomas Mensink, “Improving the Fisher Kernel for Large-Scale Image Classification”, European Conference on Computer Vision, pp. 143-156, 2010.
[Reference 5] Herve Jegou, Florent Perronnin, Matthijs Douze, Jorge Sanchez, Patrick Perez, Cordelia Schmid, “Aggregating Local Image Descriptors into Compact Codes”, IEEE Trans. Pattern Recognition and Machine Intelligence, Vol. 34, No. 9, pp. 1704-1716, 2012.

  In any case, the resulting feature quantity is a real value vector having a length that depends on the number of codes of the code length.

  A concept feature is an object included in an image or an event captured by the image. Anything may be used, but examples include “sea”, “mountain”, “ball”, and the like. If “sea” appears in an image, the image belongs to the “sea” concept. Whether or not the image belongs to each concept can be determined using a concept classifier. Usually, one concept discriminator is prepared for each concept, and the feature amount of the image is input, and whether or not the image belongs to a certain concept is output as an attribution level. The concept classifier is learned and acquired in advance, and it is a correct answer that expresses the predetermined image features, for example, the local features described above and the concept to which the image belongs by hand in advance. Earn by learning the relationship with the label. For example, a support vector machine may be used as the learning device. Concept features can be obtained by expressing the attribution levels for each concept together as a vector. In this case, the generated feature quantity is a vector having the same number of dimensions as the number of concepts.

  A landscape feature is a feature amount that represents a landscape or scene of an image. For example, the GIST descriptor described in Reference 6 can be used. The GIST descriptor is represented by a coefficient when an image is divided into regions and a filter having a certain orientation is applied to each region. In this case, the generated feature amount is the type of filter (region to be divided). And the number of orientations).

  [Reference 6] A. Oliva and A. Torralba, “Building the gist of a scene: the role of global image features in recognition”, Progress in Brain Research, 155, pp. 23-36, 2006

  The shape feature is a feature amount representing the shape of an object shown in an image. For example, the HOG feature amount and edge histogram described in Reference 7 can be used.

  [Reference 7] N. Dalal and B. Triggs, “Histograms of Oriented Gradients for Human Detection”, IEEE Conference on Computer Vision and Pattern Recognition, pp.886-893, 2005

  When the content is a sound, a pitch feature, a sound pressure feature, a spectrum feature, a rhythm feature, an utterance feature, a music feature, a sound event feature, and the like are extracted. The pitch feature may be a pitch, for example, and can be extracted using a method described in Reference Document 8 below. In this case, the pitch may be expressed as a one-dimensional vector (scalar) or may be quantized into several dimensions.

  [Reference 8] Sadahiro Furui, “Digital Audio Processing, 4.9 Pitch Extraction”, pp. 57-59, 1985

  As the sound pressure feature, an amplitude value of speech waveform data may be used, or a short-time power spectrum may be obtained, and an average power in an arbitrary band may be calculated and used. In any case, the length vector depends on the number of bands for calculating the sound pressure.

  As the spectrum feature, for example, Mel-Frequency Cepstral Coefficients (MFCC) can be used.

  As the rhythm feature, for example, a tempo may be extracted. In order to extract the tempo, for example, a method described in Reference Document 9 below can be used.

  [Reference 9] E.D. Scheirer, “Tempo and Beat Analysis of Acoustic Musical Signals”, Journal of Acoustic Society America, Vol. 103, Issue 1, pp.588-601, 1998

  The utterance feature and the music feature represent the presence or absence of utterance and the presence or absence of music, respectively. In order to find a section where speech / music exists, for example, a method described in Reference Document 10 below may be used.

  [Reference 10] K. Minami, A. Akutsu, H. Hamada, and Y. Tonomura, “Video Handling with Music and Speech Detection”, IEEE Multimedia, vol. 5, no. 3, pp. 17-25, 1998

  As the sound event feature, for example, emotional sound such as laughter and loud voice, or occurrence of environmental sound such as gunshot and explosion sound may be used. In order to detect such a sound event, for example, a method described in Reference Document 11 below may be used.

  [Reference 11] International Publication No. 2008/032787

  When the content is a video, since the video is generally a stream of images and sounds, the image features and sound features described above can be used. As to which image and sound information in the video is analyzed, for example, the video is divided into several sections in advance, and feature extraction is performed from one image and sound for each section.

  In order to divide the video into sections, the video may be divided at predetermined intervals, for example, by using the method described in Reference Document 12 below or the like, where the video is cut discontinuously. It is good also as what divides | segments by the cut point which is.

  [Reference 12] Y. Tonomura, A. Akutsu, Y. Taniguchi, and G. Suzuki, “Structured Video Computing”, IEEE Multimedia, pp.34-43, 1994

  When the video is divided into sections, the latter method is desirably employed. As a result of the video section division process, the start point (start time) and end point (end time) of the section are obtained, and may be handled as separate feature quantities at each time.

  One or a plurality of feature quantities described above may be used, or other known feature quantities may be used.

  The hash function generation unit 32 performs a hash based on the feature amount of each of the N pieces of content extracted by the feature extraction unit 30 and the semantic vector provided for each of the N pieces of content acquired by the input unit 2. Generate a function. The hash function is a hash function for obtaining a hash value corresponding to the feature amount.

Specifically, when the feature quantity extracted from a certain content i is represented as x _i ∈R ^D and the feature quantity dimension of the content of the feature quantity x _i is D, the hash function generation unit 32 determines the feature quantity x _i. In contrast, a set of hash functions such that h _k : R ^D → {−1, 1} is obtained. If Note that there is no essentially difference in terms of the amount of information and {-1, 1} and {0,1}, by each h, characteristic amounts _{x i} ∈R ^D binary value takes 0 or 1 Therefore, the feature quantity x _i is converted into B binary values, that is, B-bit hash values by the hash function set H = {h ₁ , h ₂ ,..., H _B }. Means. In the present embodiment, a case where a hash value is composed of a plurality of binary values will be described as an example.

  An object of the embodiment of the present invention is to enable measurement of similarity even with different media types using this hash value, and to omit time-consuming similarity calculation. Therefore, the hash function generated here and the hash value generated thereby have the following two properties.

(A) Conversion into a hash value representing the similarity in the original space ^RD . That is, the content having a higher similarity has a shorter hash value distance (Hamming distance).

(B) Hash values are close to each other in content having similar semantic vectors.

  In an example of the present embodiment, a hash function based on a linear function expressed by equation (1) is applied as a hash function.

Here, sign (x) is a sign function, which is 1 when x ≧ 0 and -1 when x <0. The parameters are w _k εR ^D and b _k εR. In this hash function, there are only two unknown parameters, w _k and b _k .

Here, if x _i (i = 1, 2,..., N) is normalized to an average of 0, generality is not lost even if b _k = 0. To normalize the x _i to 0, the average of _{x i,} and than can be subtracted from each _{x i,} which is because it is always possible in _{x i} ∈R _^D, it can be determined and _b k = 0 . Therefore, hereinafter, it is assumed that the average of x _i is normalized to 0, and the above equation (1) is redefined as equation (2).

According to the definition of the hash function, the hash function can be uniquely determined by determining the parameter w _k in the function φ _k constituting the hash function. Therefore, the purpose of this hash function generation process is to obtain this w _k (k = 1, 2,..., B).

Here, the meaning of the hash function defined as the above equation (2) can be geometrically explained with reference to FIG. In FIG. 2, feature amounts x _i (i = 1, 2,..., N) extracted from each content (i = 1, 2,..., N) are stored in the feature amount space ^RD. Distributed. In FIG. 2, for the sake of convenience, it is illustrated as two-dimensional, but it is actually a D-dimensional space. Here, the function φ _k (x) constituting the hash function represents a straight line (actually a D−1-dimensional hyperplane) passing through the origin on the feature amount space. Since h _k (x) is essentially a sign function, its value is 1 or 2 depending on which side of the straight line of the function φ _k (x) that constitutes the hash function the feature point is. Take 0. That is, the hash function 41 defined by the above equation (2) is a function that divides the feature amount space into two areas of 1 and 0 by a straight line. Here, w _k corresponds to the slope of this straight line, and if w _k changes, the angle to be divided will change.

In order to obtain w _k so that the hash function matches the two properties (A) and (B) described above,

(A ′) Draw a straight line so that similar content groups gather on one side of the straight line in the example of FIG. 2 (that is, determine w _k ),

(B ′) Furthermore, the related contents indicated by the semantic vector may have close values in the hash value space.

  First, a method for satisfying the property (A ′) will be described. In media content, it is well known that the distribution of feature quantities and semantic vectors between similar contents forms a smooth manifold structure regardless of the types of feature quantities described above. A manifold structure is simply a smooth change. If it demonstrates easily using FIG. 3, each feature-value will be roughly distributed on two curves of the curve 51 and the curve 52 which change smoothly, and the points on the same curve are mutually similar. There are many cases. The feature values represented by white circles (◯) and black circles (●) in FIG. 3 are content features that are similar to each other if they are the same color.

  Further, the same can be said when a semantic vector space formed by semantic vectors is considered.

Based on this knowledge, a straight line may be drawn so that similar content groups in these feature amount space or semantic vector space gather on one side of the straight line. Based on this point of view, a straight line such as the straight line 61 is not preferable among the straight lines shown in FIG. 4, and a hash function parameter w _k that defines a straight line such as the straight line 62 passing between the two groups may be obtained. become.

  Next, a method for satisfying the property (B ′) will be described with reference to FIG. In the example of FIG. 5, the feature amount space formed by the feature amount and the semantic vector space formed by the semantic vector corresponding to each content are shown in association with each other, the feature amount is a circle, and the semantic vector is a triangle. It is represented by It is assumed that straight lines 71 and 72 that satisfy the property (A ′) in each feature amount space and semantic vector space, that is, separate manifold structures, are obtained, respectively. It is assumed that the correspondence between the feature quantity and the semantic vector is obtained as indicated by the broken lines 73 to 76 (in the figure, the correspondence is shown only for three contents, but in actuality with respect to other contents) May be considered to have a similar correspondence).

At this time, if the hash function parameter w _k is calculated so that the corresponding feature quantities / semantic vectors among the feature quantities and the semantic vectors separated by the straight lines 71 and 72 have the same hash value, Similar hash values can be generated both in the meaning of the feature quantity and in the meaning of the semantic vector, and satisfy the property (B ′). For example, in the example of FIG. 5, the white circle and the white triangle (Δ), and the black circle and the black triangle (▲) may have the same hash value.

Based on the two methods described above, in the example of the present embodiment, the parameter w _k satisfying the two properties (A ′) and (B ′) described above is _obtained . In an example of this embodiment, w _k is _obtained by the following two procedures. The first procedure captures the manifold structure in each of the feature space and the semantic vector space. The second procedure obtains w _k based on the manifold structure of each of the feature amount space and the semantic vector space, and the correspondence between the contents.

  Each procedure is described in detail below. The first procedure is the same regardless of the feature quantity and the semantic vector, and the same process may be applied to both, so only the case of the feature quantity will be described here. For example, a known method described in Non-Patent Document 2 can be used. Hereinafter, the method described in Non-Patent Document 2 will be described.

  A manifold is roughly a smooth figure, in other words, it can be regarded as a Euclidean space when viewed locally. For example, it may be interpreted as being approximated as a collection of several straight lines like the curve shown in FIG. This means that a manifold has a structure that is approximated linearly when viewed locally, in other words, any point on a manifold is a number of points on the same manifold. It means that it can be expressed by the relative geometric relationship of the neighborhood based on the neighborhood point.

  In the said nonpatent literature 2, a manifold is discovered by solving the following problems.

Here, the first term is the feature quantity x _i , a set of feature quantities {x _j | j∈ε (x _i ) corresponding to the feature quantity index included in the neighborhood set ε (x _i ) in the Euclidean space. }, S _ij is a connection weight at that time. The second term requires that the elements of the weight vector s _i = {s _i1 ,..., S _{i | ε (xi) |} } be sparse, ie how many in the vector It is a sparse term that regularizes only such limited elements so as to have a non-zero value, and v _i is a vector having constants that have smaller values closer to x _i as elements. The element v _ij of the vector v _i is expressed by, for example, the following equation (4). Note that the weight s _{i = j} for its own vector is zero.

In other words, by solving this problem, it is possible to obtain a connection weight s _i when a certain feature amount x _i is expressed as a linear combination of as few neighboring points as possible. It represents the point and its relative geometric relationship (join weight). This problem can be solved by known sparse problem solvers. For example, open source software such as SPAMS may be used.

The neighborhood set ε (x _i ) may be obtained using any method. The simplest method is to obtain Euclidean distances with respect to all the other points x _{j ≠ i} for each feature quantity x _j , and use t items from the closest to the neighborhood set. t may be any positive integer, for example, t = 10.

However, in this method, since it is necessary to obtain the distance from one feature quantity to all the other feature quantities, if an attempt is made to obtain a neighborhood set for an unknown feature quantity x _j , the calculation of O (N) is performed. There is a problem that it takes time. Therefore, it is preferable to use a method capable of calculating at high speed. For example, clustering or hashing methods can be used.

When clustering is used, for example, all N feature quantities are clustered by the k-means method or the like, and L clusters (L << N) and L representative feature quantities (cluster centers) representing each cluster are obtained. I ask for it. The value of L may be an arbitrary positive integer, but may be L = 128, for example. As a result, it is possible to obtain which cluster each feature quantity belongs to and a representative feature quantity of the cluster. Under this premise, a neighborhood set for an unknown feature quantity x _i can be obtained by the following procedure. First, the feature amount x _j, calculates the distance between the L representative characteristic quantity, identifying the nearest cluster. Next, all feature quantities belonging to the cluster are obtained as a neighborhood set ε (x _i ). Since the calculation time required for this processing is O (L) and L << N, a neighborhood set can be obtained at a higher speed than in a simple method.

When using a hash, for example, the hash values for all N feature quantities are obtained by the method described in Non-Patent Document 1, for example. Under this assumption, the hash value of the unknown feature quantity x _j is obtained, and all the feature quantities having the same hash value or a value close to the Hamming distance (that is, belonging to the same or close bucket) are obtained. And a neighborhood set ε (x _i ). The calculation time required for this process depends on the number of buckets to be referenced, but since the number of reference buckets is generally smaller than N, a neighborhood set can be obtained at high speed as well. Note that the hash value by the method of Non-Patent Document 1 is a hash function that preserves the cosine similarity in the Euclidean space, and the closer the angle in the Euclidean space is, the higher the probability that the hash values collide. On the other hand, the hash value generated according to the present embodiment is a hash function that stores the proximity on the manifold (closeness based on the geodesic distance), not on the Euclidean space, and the distribution of the feature amount is A more accurate captured hash value can be generated.

  The above procedure may be applied to both feature quantities and semantic vectors.

Next, the second procedure will be described. W _k is obtained by obtaining a hash space (embedding) having a relative geometric relationship in the vicinity similar to s _i (i = 1, 2,..., N) of the feature amount obtained by the first procedure. .

Hereinafter, the feature quantity of the i-th content is represented as x _i and the semantic vector is represented as y _i as before.

Specifically, the following problems are solved. For convenience, a matrix X = {x ₁ ,... In which feature amounts x _i (i = 1, 2,..., N) and semantic vectors y _i (i = 1, 2,..., N) are arranged. , X _N }, Y = {y ₁ ,..., Y _N }. Furthermore, in addition to the hash function parameters w _k (k = 1, 2,..., B), pseudo hash function parameters u _k (k = 1, 2,..., B) for the semantic vectors are introduced. Define a matrix W = {w ₁ ,..., W _B }, U = {u ₁ ,..., U _B }.

  Solve the following problems:

Here, S _X / S _Y is a matrix having s _ij for each feature quantity / semantic vector (s _ij is obtained by the first procedure).

J _X and J _Y are functions for saving the manifold structure in the space of the feature value and the semantic vector, respectively, and can be defined as, for example, Expression (6).

The manifold structure in the above formula (6) is a hash that corresponds to the feature amount of the content in the feature amount space, which is the space where the feature amount exists, corresponding to the feature amount of the other content existing in the vicinity of the feature amount of the content. It is a linear combination of values.

The above equation (6) is a function that constitutes a hash function by using a manifold structure in the original feature space, that is, s _ij and its linear combination.

It is requested to reconstruct as it is even in the destination converted by the above, and is similar to the above equation (3). That is, when assigned to the above equation (5), W can be determined so as to have a manifold structure similar to the original space even after being converted into a hash value. The above has been described an example of a J _X, may be exactly the same applies to the J _Y. The equation for J _Y shown in equation (8).

  The manifold structure in the above equation (8) is a hash corresponding to the semantic vector of another content existing in the vicinity of the semantic vector of the content in the semantic vector space where the semantic vector exists. It is a linear combination of values.

J _XY in the above (5) is a function for saving the relationship between the mean vector and the feature quantity of the same content, for example, can be defined as equation (9).

  In this equation, the feature amount and the semantic vector of the same content are requested to have the same value at the conversion destination. The above expression represents the distance between values converted by the functions constituting the hash function given by the above expression (7) for each pair of feature quantity and semantic vector. Therefore, by substituting this into the above equation (5), W (and U) can be determined so as to make the distance as small as possible.

  Therefore, the hash function generation unit 32 corresponds to the feature amount extracted from the content for each of the N pieces of content according to the above formulas (5), (6), (8), and (9). A hash value corresponding to a feature amount of the content obtained based on a manifold structure for each of the N contents, and a distance between the hash value and a hash value corresponding to the semantic vector of the content is reduced; A hash value corresponding to a semantic vector of the content obtained based on the manifold structure for each of the N content, and a distance from the hash value corresponding to the feature value extracted from the content is reduced. The hash value corresponding to the feature value is obtained so that the distance from the hash value corresponding to the semantic vector of the content is reduced. Generating a hash function for obtaining a hash value corresponding to the hash function, and means a vector of order.

  Substituting the above defined equations (6), (8) and (9) into the above equation (5) and applying algebraic deformation, the following problem of equation (10) is obtained. .

  here,

It is. Since the above expression (10) is convex with respect to W (and U), that is, P, differentiation with respect to W and taking an extreme value results in the following generalized eigenvalue problem.

  In the above equation (12), η represents an eigenvalue. A solution of such a generalized eigenvalue problem can be obtained by a known method such as an iterative method or a power method.

  W (and U) obtained in this way optimally preserves the manifold structure in the original space, and converts the relationship between the same content and the semantic vector to the same hash value as much as possible. . Therefore, it is possible to obtain W that optimally satisfies the two properties (A ′) and (B ′) that were intended.

When generating a new hash value, it is only necessary to check the sign after calculating the above equation (7). Amount of memory required for this calculation is only the amount of memory required to store each w _k and x _i, if a feature quantity is a floating-point representation, if the dimension D is 100 800B about, if the dimension Even if D becomes about 100,000, it can be suppressed to 800 KB at most, and the amount of memory that can be stored extremely easily even in an existing general computer. Therefore, this method enables high-speed and memory-saving hash value generation with high accuracy by capturing the manifold structure.

  The hash function generation unit 32 is based on the feature amount of each of the N pieces of content extracted by the feature extraction unit 30 and the content semantic vector given to each of the N pieces of content received by the input unit 2. In accordance with the equations (10), (11), and (12), a hash function for obtaining a hash value corresponding to the feature value and a hash function for obtaining a hash value corresponding to the semantic vector are generated. .

  The hash function storage unit 34 stores a hash function for obtaining a hash value corresponding to the feature amount generated by the hash function generation unit 32. A hash function for obtaining a hash value corresponding to the feature amount generated by the above processing details, that is, specifically, W for all media types is stored in the hash function storage unit 34.

For each of the N pieces of content stored in the content database 5, the hash value generation unit 36 is based on the feature amount extracted by the feature extraction unit 30 and the hash function stored in the hash function storage unit 34. A hash value corresponding to the feature amount of the content is generated.
In addition, the hash value generation unit 36 uses the feature amount extracted by the feature extraction unit 30 and the hash function stored in the hash function storage unit 34 for the content as a search query received by the input unit 2. A hash value corresponding to the feature amount of the content is generated.

  The output unit 4 outputs the hash value generated by the hash value generation unit 36 to the content database 5.

  The content database 5 stores a combination of the hash value output by the output unit 4 and the content corresponding to the hash value.

<Operation of information processing device>
Next, the operation of the information processing apparatus 1 according to this embodiment will be described. The information processing apparatus 1 executes a hash function generation process for generating a hash function and a hash value generation process for hashing the feature amount. Hereinafter, these two processes will be described.

<Hash function generation processing routine>
First, when the information processing apparatus 1 acquires N contents stored in the content database 5 and N meaning vectors stored in the semantic vector database 6, the information processing apparatus 1 uses the hash shown in FIG. A function generation processing routine is executed. The hash function generation processing routine is a process executed at least once before the content data is actually hashed.

  First, in step S100, the input unit 2 acquires a plurality (N) of contents stored in the content database 5 and N semantic vectors stored in the semantic vector database 6.

  In step S102, the feature extraction unit 30 extracts a feature amount from the content for each of the N pieces of content acquired in step S100.

  In step S104, the hash function generation unit 32 based on the feature amount of each of the N contents extracted in step S102 and the semantic vector given to each of the N contents acquired in step S100. Thus, a hash function for obtaining a hash value corresponding to the feature amount and a hash function for obtaining a hash value corresponding to the semantic vector are generated according to the above equations (10) to (12). Further, the hash function generation unit 32 stores the hash function for obtaining the hash value corresponding to the feature amount and the hash function for obtaining the hash value corresponding to the semantic vector in the hash function storage unit 34, and The function generation processing routine is terminated.

  Through the above processing, a hash function can be generated from the N contents stored in the content database 5 and the N semantic vectors stored in the semantic vector database 6.

<Hash value generation processing routine>
Next, when content as a search query is input to the information processing apparatus 1, the information processing apparatus 1 executes a hash value generation processing routine shown in FIG. 7. The hash value generation processing routine is a process of hashing the content feature amount using a hash function for obtaining a hash value corresponding to the feature amount stored in the hash function storage unit 34.

  First, in step S200, the input unit 2 accepts input of content as a search query.

  In step S202, the feature extraction unit 30 extracts feature amounts from the content received in step S200.

  In step S204, based on the feature value extracted in step S202 by the hash value generation unit 36 and the hash function stored in the hash function storage unit 34 in step S104, the content received in step S200 is stored. Generate a hash value. Further, for each of the N pieces of content stored in the content database 5 by the hash value generation unit 36, each feature amount of the content extracted in step S102 and the hash function storage unit 34 in step S104 are stored. Based on the stored hash function, a hash value of the content is generated.

  In step S206, the output unit 4 determines a combination of the hash value generated in step S204 and the content received in step S200, and a combination of the hash value generated in step S204 and the N content. The data is output to the content database 5, stored in the content database 5, and the hash value generation processing routine is terminated.

  Through the above processing, a hash value can be obtained for the input content regardless of the type of media. In addition, content having a hash value similar to the hash value of content as a search query can be found from N pieces of content.

  As described above, according to the information processing apparatus according to the embodiment of the present invention, based on the feature amount of each of the plurality of contents and the semantic vector representing the meaning of the content given to each of the plurality of contents. Then, for each of the plurality of contents, the hash value corresponding to the feature value is reduced so that the distance between the hash value corresponding to the feature value extracted from the content and the hash value corresponding to the semantic vector of the content is reduced. By generating a hash function for obtaining the hash value and a hash function for obtaining the hash value corresponding to the semantic vector, a hash function for obtaining a hash value in consideration of the meaning of the content can be generated.

  In addition, according to the information processing apparatus according to the embodiment of the present invention, by generating a hash value corresponding to the feature amount of the content based on the feature amount extracted from the content and the generated hash function, A hash value considering the meaning of the content can be generated. In addition, based on the feature amount extracted from the content as the search query and the generated hash function, the hash value corresponding to the feature amount of the content as the search query is generated, thereby taking into account the meaning of the content. Using the hash value, content similar to the content as the search query can be found with high accuracy.

  Further, as described above, according to the present embodiment, a semantic viewpoint can be obtained by generating a parametric hash function so as to capture the manifold structure of the feature amount space and preserve the closeness of the semantic vector. Similar contents can be found with high accuracy while being fast and memory-saving.

  Also, similar content can be found from a large amount of content.

  In addition, the content can be found with high accuracy while considering various semantic viewpoints while being high speed and saving memory.

<Second Embodiment>
<System configuration>
Next, a second embodiment of the present invention will be described. In addition, about the part which becomes the structure similar to 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  In the second embodiment, the type of hash function is different from that in the first embodiment.

In the second procedure described above in the first embodiment, in the case of a hash function taking the form of the above formula (2), a method for obtaining the parameter w _k (k = 1, 2,..., B) As described above, the hash function that can be handled in the embodiment of the present invention is not limited to this form, and even if the hash function takes another form, its parameters can be similarly determined.

  For example, the following hash function can be handled.

Here, α _k , _t is a parameter, and κ (x _t , x) is a kernel function. The kernel function

And for N feature quantities {x ₁ ,..., X _N },

And for any real number α _i , α _j

Any function that satisfies There are an infinite number of such functions.

Etc. exist. However, β and γ are positive real value parameters, and p is an integer parameter, which may be determined as appropriate.

In the above equation (13), b _k is

That is, since the constant is determined by the average value, the above equation (13) is

And can be converted to the inner product form. However,

It is. Here, T is a constant that determines the hash function. The kernel function constituting the hash function, specifically the kernel vector map κ (x), is determined by T feature quantities, but T may be set to an arbitrary value in the range of T <N. For example, assuming that T = 300, T may be selected at random from all the feature quantities {x ₁ ,..., X _N }, or a representative vector selected using a clustering method such as K-means. It is good. In the present embodiment, a case where T feature amounts are selected in advance will be described as an example.

  A hash function defined in this way can handle a nonlinear mapping defined in the form of a kernel function. Therefore, it has an advantage that it can be expressed more flexibly than the hash function according to the above equation (6) in that it can handle not only a non-linear function, that is, a straight line but also a curve.

  The hash function generation unit 32 generates a kernel function based on the feature amounts of T pieces of content selected in advance.

Hereinafter, a method of determining the parameter α _k in the hash function taking the form of the equation (13) will be described. For convenience, a matrix た_X == κ _X (x _i ) (i = 1, 2,..., N) for feature quantities and κ _Y (y _i ) (i = 1, 2,..., N) for semantic vectors. {Κ _X (x ₁ ),..., Κ _X (x _N )}, Κ _Y = {κ _Y (y ₁ ),..., Κ _Y (y _N )} are defined. Further, a hash function parameter α _k (k = 1, 2,..., B) for the feature quantity and a pseudo hash function parameter θ _k (k = 1, 2,..., B) for the semantic vector. Α = {α ₁ ,..., Α _B } and Θ = {θ ₁ ,..., Θ _B } are defined.

  Specifically, the following problem corresponding to the above equation (5), which is the hash function defined by the above equation (2), is solved.

J _X , J _Y and J _XY can be defined, for example, as follows for the same reason as the above formulas (6), (8), and (9).

Substituting the above equations (20), (21), and (22) into the above equation (19) and applying algebraic deformation, the following problem of equation (23) is obtained.

  here,

It is. This problem is equivalent to the problem of the above equation (10), and can be solved by exactly the same procedure.

The hash function generation unit 32 is based on the feature amount of each of the N pieces of content extracted by the feature extraction unit 30 and the content semantic vector given to each of the N pieces of content received by the input unit 2. According to the above equations (23) and (24), for each of the N contents, the distance between the hash value corresponding to the feature amount extracted from the content and the hash value corresponding to the semantic vector of the content is For each of the N pieces of content, the distance between the hash value corresponding to the feature amount of the content obtained based on the manifold structure and the hash value corresponding to the feature amount extracted from the content is For each of the N pieces of content, the content of the content is calculated based on the manifold structure. A hash function for obtaining a hash value corresponding to a feature value and a hash value corresponding to a semantic vector so that the distance between the hash value corresponding to the taste vector and the hash value corresponding to the semantic vector of the content is reduced. Generate a hash function to find
A hash function and a kernel function for obtaining a hash value corresponding to the feature amount generated by the hash function generation unit 32 are stored in the hash function storage unit 34. Specifically, the parameter Α and the kernel function κ (x) are stored in the hash function storage unit 34.

For each of the N contents stored in the content database 5, the hash value generation unit 36 converts the feature amount extracted by the feature extraction unit 30 and the hash function and kernel function stored in the hash function storage unit 34. Based on this, a hash value corresponding to the feature amount of the content is generated.
In addition, the hash value generation unit 36 divides the content as a search query received by the input unit 2 into the feature amount extracted by the feature extraction unit 30 and the hash function and kernel function stored in the hash function storage unit 34. Based on this, a hash value corresponding to the feature amount of the content is generated.

  Note that other configurations and operations of the information processing apparatus according to the second embodiment are the same as those of the first embodiment, and thus description thereof is omitted.

<Third Embodiment>
<System configuration>
Next, a third embodiment of the present invention will be described with reference to FIG. In addition, about the part which becomes the structure similar to 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  The third embodiment is different from the first and second embodiments in that an information processing system is configured by a server device and a client device. In the third embodiment, a case where the present invention is applied to an information processing system that performs similar content search will be described as an example.

  An information processing system 200 according to the third embodiment of the present invention includes a server device 7 and a client device 13 as shown in FIG.

  The server device 7 shown in FIG. 8 is configured by a computer including a CPU, a RAM, and a ROM that stores a program for executing each processing routine, and is functionally configured as follows. . As shown in FIG. 8, the server device 7 includes an input unit 8, a calculation unit 9, and an output unit 10. The calculation unit 9 includes a feature extraction unit 90, a hash function generation unit 92, a hash function storage unit 94, and a hash value generation unit 96. Further, a plurality of contents are registered in the content database 11 as in the case of the content database 5 of the first embodiment. The semantic vector database 12 stores semantic vectors corresponding to the respective contents registered in the content database 11 as in the semantic vector database 6 of the first embodiment.

  Further, the client device 13 shown in FIG. 8 is configured by a computer including a CPU, a RAM, and a ROM storing a program for executing each processing routine, and is functionally configured as follows. ing. As shown in FIG. 8, the client device 13 includes an input unit 14, a calculation unit 15, and an output unit 16. The calculation unit 15 includes a feature extraction unit 150, a hash function storage unit 154, and a hash value generation unit 156.

  Here, in the server device 7 and the client device 13, the common components (input unit, feature extraction unit, hash function storage unit, hash value generation unit) are configured to have the same functions, respectively, Those having the same names as the constituent elements described in FIG. 1 may have the same functions as those in FIG. Furthermore, it is assumed that the contents of the hash value generation unit are appropriately synchronized by some communication means.

  Processing operations in the apparatus configuration shown in FIG. 8 are as follows. First, the server device 7 performs processing similar to the processing described above, generates a hash function as appropriate, stores it in the hash function storage unit 94, and synchronizes with the hash function storage unit 154 of the client device 13. Further, a process similar to the process described above is performed on the content in the content database 11 to generate a hash value and store it in the content database 11.

  On the other hand, when the client device 13 receives a search request from a user, that is, an input of content as a search query, by the input unit 14, the client device 13 generates a hash value for the content, and the output unit 16 sends the server device 7. The hash value is output to the input unit 8.

  When the hash value is received from the client device 13, the server device 7 searches the content database 11 using the hash value, finds similar content based on the hash value, and uses the result as the client device 13. To output.

  Finally, the client device 13 outputs the search result received from the server device 7 to the user.

  With this configuration, the server device 7 can perform the hash function generation process, and the client device 13 can perform only the hash value generation process.

  Note that the other configuration and operation of the information processing system 200 according to the third embodiment are the same as those in the first embodiment, and thus the description thereof is omitted.

  The merit of taking this configuration will be described. In general, a client device (PC, portable terminal, etc.) has poor calculation capability compared to a server device, and therefore may not be suitable for processing with a relatively large amount of calculation such as hash function generation. According to this configuration, the hash function generation process can be appropriately performed by a server apparatus having high calculation capability, and only the hash value generation process with a small calculation amount can be performed by the client apparatus. In addition, when transmitting information with a large amount of data by communication via a network, there is a problem that it takes a long transmission time. However, according to the configuration, only a plurality of feature amount hash values having a small amount of information may be transmitted. Thus, the quick response to the search can be improved.

  As explained above, by using the hash function and hash value described above, it is possible to find similar contents including various semantic viewpoints and their strengths with high accuracy while being fast and memory-saving. become. With this configuration, since the memory is saved, for example, the mobile terminal (smart phone or tablet) having a small memory amount can be used. Moreover, since it is high-speed, it can respond also to the use for which real-time property is required. As specific usage scenes utilizing these effects, it is possible to take a picture of a place or product that is of interest when walking in the city with a mobile terminal and search for a similar place or product.

[Example]
Next, an example of an embodiment that searches for similar contents in a high-speed and memory-saving manner using the hash function generated by the processing described in the first embodiment will be described. If the hash function generation processing described above has been completed, the hash function storage unit 34 stores B sets of hash functions for each media type. If this is used, any content expressed by the feature value can be expressed by a hash value of B bits or less according to the above equation (2). For example, the content database 5, the N image content is stored, the feature quantity _{X = {x 1, ···,} x N} and the corresponding mean vector _{Y = {y 1, ···,} y It is assumed that _N } is stored in the semantic vector database 6, and all feature quantities are converted into hash values Z = {z ₁ ,..., Z _N } based on the above equation (2). At this time, an object is to find content similar to the feature quantity x _q not included in X from X.

First, the feature quantity x _q is converted into the hash value z _q based on the above equation (2). The simplest method is to use a hash table shown in FIG. First, a hash table as shown in FIG. 9 is configured by the hash value Z registered in the content database 5. In this table, a hash value and a feature value (content identifier) converted to the hash value are stored in association with each other. When a hash value is given, the content having the same hash value is immediately stored. Can be found. Here, the hash values generated by the embodiment of the present invention are characterized in that related and similar ones can be converted into the same hash value regardless of the media type. That is, for example, when the hash value “0000” is specified, the image (image 1, image 3...) Associated with the hash value “0000” can be immediately found without depending on the media type. Similarly, given the benefit of this hash table, it is possible to find instantly content corresponding to the hash value z _q.

  According to this method, the number of image contents registered in the content database 5 does not depend on the number N, and it is not necessary to store the original feature amount in the memory at a high speed in a substantially constant time. There is an advantage that can be found.

As another method, distance calculation based on the Hamming distance can be used. That is, the distance between the hash value z _q and the N hash values included in Z is calculated, and the one with a small distance is obtained as similar content. Since the hash value is binary, the distance calculation can be performed by, for example, the Hamming distance, but the Hamming distance is calculated by XOR (exclusive OR) and popcnt operation (that is, the bit of 1 in the binary string). The calculation can be performed only by counting the number), and the hash value can usually be expressed by a small number of binary values. Therefore, the calculation can be performed much faster than the distance calculation using the original feature amount.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention. Needless to say, any application and configuration can be adopted within a range that satisfies the main features of the present embodiment.

  In addition, the hash function generation unit 32 and the hash value generation unit 36 in the first embodiment may be configured as separate devices. In that case, a hash function generation device is configured including the hash function generation unit 32, and a hash value generation device is configured including the hash value generation unit 36.

  The information processing apparatus and the information processing system according to the present embodiment have been described with respect to the case where the hash function storage unit 34 is provided. For example, the hash function storage unit 34 is provided in the information processing apparatus and an external device of the information processing system. The information processing apparatus and the information processing system may refer to the hash function storage unit 34 by communicating with an external apparatus using a communication unit.

In the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium.
For example, the hash function generation unit and the hash value generation unit in the above-described embodiment may be realized by a computer. In that case, a program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be read into a computer system and executed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, the “computer-readable recording medium” dynamically holds a program for a short time like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, a volatile memory inside a computer system serving as a server or a client in that case may be included and a program held for a certain period of time. Further, the program may be for realizing a part of the functions described above, and may be a program capable of realizing the functions described above in combination with a program already recorded in the computer system. It may be realized using hardware such as PLD (Programmable Logic Device) or FPGA (Field Programmable Gate Array).

  In addition, the information processing apparatus and the information processing system described above have a computer system inside, but if the “computer system” uses a WWW system, a homepage providing environment (or display environment) Shall also be included.

  As mentioned above, although embodiment of this invention has been described with reference to drawings, the said embodiment is only the illustration of this invention, and it is clear that this invention is not limited to the said embodiment. is there. Therefore, additions, omissions, substitutions, and other modifications of the components may be made without departing from the technical idea and scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Information processing apparatus 2,8,14 Input part 3,9,15 Calculation part 4,10,16 Output part 5,11 Content database 6,12 Meaning vector database 7 Server apparatus 13 Client apparatus 30,90,150 Feature extraction part 32, 92 Hash function generators 34, 94, 154 Hash function storage units 36, 96, 156 Hash value generator 200 Information processing system

Claims (8)

A hash function generation method in a hash function generation device including a feature extraction unit and a hash function generation unit,
The feature extraction means for extracting a feature amount from the content for each of a plurality of content;
The hash function generation means is based on the feature amount of each of the plurality of contents extracted by the feature extraction means and a semantic vector representing the meaning of the content given to each of the plurality of contents, For each of the plurality of contents, a hash value corresponding to the feature amount so that a distance between a hash value corresponding to the feature amount extracted from the content and a hash value corresponding to the semantic vector of the content is reduced. Generating a hash function for obtaining a hash value for obtaining a hash value corresponding to the semantic vector;
Hash function generation method including The step of generating the hash function by the hash function generating means includes the feature amount of each of the plurality of contents extracted by the feature extracting means and the meaning of the content given to each of the plurality of contents. For each of the plurality of contents, the feature quantity of the content is present in the vicinity of the feature quantity of the content in a feature quantity space in which the feature quantity exists for each of the plurality of contents. A hash value corresponding to the feature amount of the content obtained based on a manifold structure represented by a linear combination of hash values corresponding to the feature amount of the content, and a hash corresponding to the feature amount extracted from the content The distance to the value is reduced, and each of the plurality of contents includes the content. A hash function for obtaining a hash value corresponding to the feature value so that a distance between a hash value corresponding to the feature value extracted from the hash value and a hash value corresponding to the semantic vector of the content is reduced, and The hash function generation method according to claim 1, wherein a hash function for obtaining a hash value corresponding to the semantic vector is generated. The step of generating the hash function by the hash function generating means includes the feature amount of each of the plurality of contents extracted by the feature extracting means and the meaning of the content given to each of the plurality of contents. For each of the plurality of contents, the meaning vector of the content is present in the vicinity of the meaning vector of the content in the meaning vector space in which the meaning vector exists. A hash value corresponding to the semantic vector of the content obtained based on a manifold structure represented by a linear combination of hash values corresponding to the semantic vector of the content, and a hash value corresponding to the semantic vector of the content The distance is reduced and the plurality of containers are In order to obtain the hash value corresponding to the feature amount so that the distance between the hash value corresponding to the feature amount extracted from the content and the hash value corresponding to the semantic vector of the content is reduced The hash function generation method of Claim 1 or Claim 2 which produces | generates the hash function for calculating | requiring the hash function corresponding to the said hash function and the said semantic vector.   The hash function generation method according to any one of claims 1 to 3, wherein the hash value is a binary value. A hash value generation method in a hash value generation device including a feature extraction unit and a hash value generation unit,
The feature extracting means extracting a feature quantity from the content;
The hash value corresponding to the feature amount generated by the hash value generation method according to any one of claims 1 to 3, wherein the hash value generation unit extracts the feature amount extracted by the feature extraction unit. Generating a hash value corresponding to the feature amount of the content based on the hash function for obtaining
Hash value generation method including For each of a plurality of contents, feature extraction means for extracting a feature amount from the contents;
For each of the plurality of contents, based on the feature amount of each of the plurality of contents extracted by the feature extraction unit and a semantic vector representing the meaning of the content given to each of the plurality of contents. A hash function for obtaining a hash value corresponding to the feature amount so that a distance between a hash value corresponding to the feature amount extracted from the content and a hash value corresponding to the semantic vector of the content is reduced. And a hash function generating means for generating a hash function for obtaining a hash value corresponding to the semantic vector,
Hash function generator including Feature extraction means for extracting feature quantities from the content;
The feature of the content based on the feature amount extracted by the feature extraction unit and the hash function for obtaining a hash value corresponding to the feature amount generated by the hash function generation device according to claim 6. Hash value generation means for generating a hash value corresponding to the quantity;
Hash value generator including   The program for making a computer perform each step of the hash function production | generation method of any one of Claims 1-4, or the hash value production | generation method of Claim 5.