CN111177438A - Image characteristic value searching method and device, electronic equipment and storage medium - Google Patents
Image characteristic value searching method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN111177438A CN111177438A CN201811341714.3A CN201811341714A CN111177438A CN 111177438 A CN111177438 A CN 111177438A CN 201811341714 A CN201811341714 A CN 201811341714A CN 111177438 A CN111177438 A CN 111177438A
- Authority
- CN
- China
- Prior art keywords
- sub
- searched
- characteristic value
- query
- image characteristic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/53—Querying
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention provides a method and a device for searching an image characteristic value, electronic equipment and a storage medium, wherein the method comprises the following steps: when a query vector used for searching a target image characteristic value matched with the sample image characteristic value is obtained, segmenting the query vector to obtain a plurality of query sub-vectors; calculating the distance from each inquiry subvector in the inquiry subvectors to all the clustering centers in the subspace corresponding to the inquiry subvectors; acquiring an asymmetric distance between a sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all clustering centers in a subspace corresponding to the inquiry sub-vector; and according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the images to be searched to obtain the characteristic values of the target images. The embodiment of the invention can improve the searching speed of the image characteristic value under the condition that the data volume is continuously increased.
Description
Technical Field
The present invention relates to the field of image search technologies, and in particular, to a method and an apparatus for searching an image feature value, an electronic device, and a storage medium.
Background
With the popularization of the internet, multimedia resources are presented on the internet in an explosive growth posture, which undoubtedly provides convenience for vigorous data search requirements, but related data are rapidly searched in billions of data volumes, and the problem which is urgently desired to be solved in the field of image search still exists. Currently, a new design proposed by Facebook artificial intelligence research team realizes a faster nearest neighbor search than the previous best GPU (Graphics Processing Unit) method, which is a similarity search tool, Faiss, a design more optimized than brute force computation, approximation and compressed domain search based on product quantization, and will be applied to different similarity search scenarios. Although the similarity search tool Faiss is today the best library of similarity search algorithms, as the amount of data continues to increase, for example: the image characteristic value data reaches the billion level or even the billion level, and the searching speed of the similarity searching tool Faiss still faces huge challenges. It is seen that, in the current image feature value search technology, the search speed is still not fast due to the rapid expansion of the data volume.
Disclosure of Invention
The invention mainly aims to provide a method and a device for searching an image characteristic value, electronic equipment and a storage medium, which are used for solving the problem that the searching speed is still not fast due to rapid expansion of data volume in the conventional image characteristic value searching technology.
In order to achieve the above object, a first aspect of the embodiments of the present invention provides a method for searching an image feature value, including:
when a query vector used for searching a target image characteristic value matched with a sample image characteristic value is obtained, segmenting the query vector to obtain a plurality of query sub-vectors, wherein the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one by one;
calculating the distance from each inquiry sub-vector in the inquiry sub-vectors to all clustering centers in the sub-space corresponding to the inquiry sub-vector, wherein the clustering centers are obtained by clustering sub-segments of a plurality of image characteristic values to be searched existing in the sub-space;
acquiring an asymmetric distance between the sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all the clustering centers in the sub-space corresponding to the inquiry sub-vector;
and according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the images to be searched, and acquiring the characteristic value of the target image.
Optionally, the step of obtaining an asymmetric distance between the sample image feature value and each image feature value to be searched according to a distance from each of the plurality of query subvectors to all cluster centers in a subspace corresponding to the query subvectors includes:
respectively aiming at each image characteristic value to be searched in the plurality of image characteristic values to be searched, the following operations are carried out:
determining the clustering center of each sub-section in the plurality of sub-sections of the image characteristic value to be searched in the sub-space of the sub-section;
determining the distance from each query subvector in the plurality of query subvectors to the cluster center of each subsegment in the subspace of the subsegment;
and performing summation operation on all the determined distances to obtain the asymmetric distance between the sample image characteristic value and the image characteristic value to be searched.
Optionally, before the step of calculating the distance from each query subvector in the plurality of query subvectors to all the cluster centers in the subspace corresponding to the query subvector, the method further includes:
respectively segmenting the image characteristic value to be searched aiming at each image characteristic value to be searched in the image characteristic values to be searched to obtain a plurality of subsections of the image characteristic value to be searched; wherein the plurality of subsections correspond to the plurality of subspaces one to one;
for each subspace of the plurality of subspaces, performing the following:
clustering all subsections existing in the subspace by adopting a clustering algorithm to obtain a plurality of clustering centers;
and taking the code of the clustering center as the quantization code of the sub-segments clustered under the clustering center, and storing the quantization codes of all the sub-segments in the plurality of index objects.
Optionally, the step of performing parallel search from a plurality of index objects used for storing the plurality of image feature values to be searched according to all the obtained asymmetric distances to obtain the target image feature value further includes:
selecting a preset number of target asymmetric distances from all the obtained asymmetric distances according to the sequence of the asymmetric distances from small to large;
performing parallel search from a plurality of index objects for storing the plurality of image characteristic values to be searched, and obtaining quantization codes of a plurality of subsegments of the image characteristic values to be searched corresponding to the target asymmetric distance;
and determining the characteristic value of the image to be searched corresponding to the target asymmetric distance according to the quantization codes of a plurality of subsections of the characteristic value of the image to be searched corresponding to the target asymmetric distance, and taking the characteristic value of the image to be searched corresponding to the target asymmetric distance as the characteristic value of the target image.
A second aspect of the embodiments of the present invention provides an apparatus for searching for an image feature value, including:
the system comprises a first segmentation module, a second segmentation module and a third segmentation module, wherein the first segmentation module is used for segmenting a query vector to obtain a plurality of query sub-vectors when the query vector used for searching a target image characteristic value matched with a sample image characteristic value is obtained, and the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one by one;
the computing module is used for computing the distance from each query subvector in the plurality of query subvectors to all the clustering centers in the subspace corresponding to the query subvectors, and the clustering centers are obtained by clustering subsections of a plurality of image characteristic values to be searched existing in the subspaces;
the acquisition module is used for acquiring the asymmetric distance between the sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all the clustering centers in the sub-space corresponding to the inquiry sub-vector;
and the searching module is used for performing parallel searching from a plurality of index objects for storing the plurality of image characteristic values to be searched according to all the acquired asymmetric distances to obtain the target image characteristic value.
A third aspect of an embodiment of the present invention provides an electronic device, including: a memory, a processor and a computer program stored on the memory and operable on the processor, the processor implementing the steps in the image feature value searching method described above when executing the computer program.
A fourth aspect of the embodiments of the present invention provides a computer-readable storage medium, which stores a computer program that, when executed by a processor, implements the steps in the above-described method for searching for an image feature value.
The scheme of the invention at least comprises the following beneficial effects:
when a query vector used for searching a target image characteristic value matched with a sample image characteristic value is obtained, segmenting the query vector to obtain a plurality of query sub-vectors, wherein the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one by one; calculating the distance from each inquiry sub-vector in the inquiry sub-vectors to all clustering centers in the sub-space corresponding to the inquiry sub-vector, wherein the clustering centers are obtained by clustering sub-segments of a plurality of image characteristic values to be searched existing in the sub-space; acquiring an asymmetric distance between the sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all the clustering centers in the sub-space corresponding to the inquiry sub-vector; and according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the images to be searched, and acquiring the characteristic value of the target image. The query vector is divided into the query sub-vectors, the query vector with higher dimensionality is actually subdivided, the segmentation is the same as the segmentation of the characteristic value of the image to be searched into a plurality of subsections in the data storage stage, and the subdivided query sub-vectors and subsections are more beneficial to improving the similarity precision of searching; the asymmetric distance is the distance between the sample image characteristic value and the quantized image characteristic value to be searched because the distance is directly calculated, so that the calculated distance can be ensured to be closer to the real distance between the sample image characteristic value and the image characteristic value to be searched, and the searching accuracy can be improved as well; the parallel search is developed from a plurality of index objects storing less data, the efficiency is much higher than that of the independent search from one index object storing a large amount of data, and the search speed of the image characteristic value can be improved even under the condition that the data amount is continuously increased.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flowchart of a method for searching an image feature value according to an embodiment of the present invention;
fig. 2 is a schematic diagram illustrating an exemplary scenario provided by an embodiment of the present invention;
fig. 3 is a schematic flowchart of another image feature value searching method according to an embodiment of the present invention;
fig. 4 is an exemplary diagram of another scenario provided by the embodiment of the present invention;
fig. 5 is a schematic structural diagram of an apparatus for searching an image feature value according to an embodiment of the present invention;
FIG. 6 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
FIG. 7 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
FIG. 8 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
FIG. 9 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
FIG. 10 is a schematic structural diagram of another apparatus for searching image feature values according to an embodiment of the present invention;
fig. 11 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The terms "comprising" and "having," and any variations thereof, as appearing in the present specification, claims and drawings, are intended to cover non-exclusive inclusions. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those steps or elements listed, but may alternatively include other steps or elements not listed, or inherent to such process, method, article, or apparatus. Furthermore, the terms "first," "second," and "third," etc. are used to distinguish between different objects and are not used to describe a particular order.
Specific embodiments of the present invention will be described below with reference to the accompanying drawings and examples, which are not intended to limit the embodiments of the present invention.
Referring to fig. 1, fig. 1 is a schematic flow chart of an image feature value searching method according to an embodiment of the present invention, as shown in fig. 1, including the following steps:
s11, when a query vector used for searching a target image characteristic value matched with the sample image characteristic value is obtained, segmenting the query vector to obtain a plurality of query sub-vectors, wherein the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one by one.
In an embodiment of the present invention, all the image feature values refer to image feature values with a higher dimension, for example: and 128 dimensions. In addition, before searching for the target image feature value, a large amount of feature value data can be adopted for pattern training to form a pattern of space segmentation, clustering and quantization coding, and finally the index object index based on product quantization is obtained.
In addition, in the feature value search stage, the sample image feature value may be a given feature value or a feature value extracted from a given image. The target image characteristic value is obtained from the prestored image characteristic value to be searched, and the search result matched with the sample image characteristic value, and the query vector used for searching the target image characteristic value can be constructed in advance or can be constructed during searching. If the query vector is obtained, according to the mode training process, firstly segmenting the query vector to obtain a plurality of query sub-vectors of the query vector, wherein the plurality of query sub-vectors correspond to the plurality of subspaces one by one. For example, the query vector is divided into 2 query sub-vectors, and then the query sub-vector No. 1 corresponds to the sub-space No. 1, and the query sub-vector No. 2 corresponds to the sub-space No. 2.
And S12, calculating the distance from each inquiry sub-vector in the inquiry sub-vectors to all the cluster centers in the sub-space corresponding to the inquiry sub-vector, wherein the cluster centers are obtained by clustering the subsections of the characteristic values of the images to be searched existing in the sub-spaces.
In an embodiment of the present invention, each of the image feature values to be searched includes a plurality of subsections having the same number as the query subvectors, and the subsections of each of the image feature values to be searched correspond to the plurality of subspaces one by one. For example, there are 2 subsections of one of the image feature values to be searched, which are No. 1 subsection and No. 2 subsection, respectively, and the corresponding No. 1 subsection correspondingly exists in the No. 1 subspace and the No. 2 subsection correspondingly exists in the No. 2 subspace. It should be noted that, since the number of feature values of the image to be searched is huge, a large number of sub-segments exist in each sub-space.
And the clustering centers are formed by clustering a large number of subsections of the characteristic values of the image to be searched in the subspace through a clustering algorithm, and the number of the clustering centers in each subspace is the same. After the query vectors are segmented according to the sample training mode, the distance from each query subvector to all the cluster centers in the subspace corresponding to the query subvector needs to be calculated. For example, the number 1 query subvector corresponds to the number 1 subspace, and all the subsections in the number 1 subspace are clustered to obtain 256 cluster centers, and then the distances from the number 1 query subvector to the 256 cluster centers need to be calculated respectively; accordingly, the query subvector # 2 also needs to calculate its distance to all cluster centers in the subspace # 2.
And S13, acquiring the asymmetric distance between the sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all the cluster centers in the sub-space corresponding to the inquiry sub-vector.
In an embodiment of the present invention, the distance from each query subvector to all the cluster centers in the subspace corresponding to the query subvector may be used to obtain an asymmetric distance between the sample image feature value and each image feature value to be searched through an operation, which may be a symmetric distance, and which operation method is specifically adopted, which is not limited herein. As shown in fig. 2, in the present invention, the distances from the query vector x to all the image feature values y to be searched actually need to be calculated, but since the number of the image feature values y to be searched may be in the order of hundreds of millions, and the number of the codewords q (y) corresponding to the index values of the image feature values y to be searched is much smaller, the distance from the query vector x to q (y) is used instead of the distance from the query vector x to the image feature values y to be searched, which is the calculation of the asymmetric distance. When the asymmetric distance is used for calculating the distance between the sample image characteristic value and the pre-stored image characteristic value to be searched, the sample image characteristic value is not required to be quantized, but the distance between the sample image characteristic value and the quantized image characteristic value to be searched is directly calculated, and the method is more direct than a symmetric distance calculation mode, can ensure that the calculated distance is closer to a real distance, and therefore improves the searching precision.
It should be noted that, the asymmetric distance or the symmetric distance between the sample image feature value and each image feature value to be searched is used to represent the similarity between the two, and a smaller distance indicates a higher similarity.
And S14, according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the image to be searched, and acquiring the characteristic value of the target image.
In an embodiment of the present invention, the index object may be pre-constructed and used to store quantization codes of a plurality of subsections of each image feature value to be searched, and since the number of the image feature values to be searched is extremely large, it is known that quantization codes of all subsections of the image feature values to be searched are also extremely large. Thus, a data threshold may be set for each index object, such as: the data threshold may be 1 hundred million, and when the quantization code stored in the initially created index object reaches the data threshold, an index object is newly created to store the quantization code that is not stored, and finally a plurality of index objects are obtained.
In addition, since all quantization codes are stored in a plurality of index objects, parallel search can be performed from the plurality of index objects according to all the acquired asymmetric distances, and the target image feature value can be obtained as a search result. For example, according to the obtained asymmetric distance, 10 image feature values to be searched are searched from the first index object, and 15 image feature values to be searched are searched from the second index object, and then the searched 25 image feature values to be searched can be used as the target image feature values.
In the method, because the parallel search is developed from the plurality of index objects according to the asymmetric distance between the calculated sample image characteristic value and each image characteristic value to be searched, the search mode is obviously much higher than the efficiency of independently searching from one index object with mass data.
It should be noted that the image feature value search method provided by the embodiment of the present invention may be implemented based on some search tools or a similarity media file search algorithm library, for example, a similarity search tool Faiss proposed by a Facebook artificial intelligence research team.
In the embodiment of the invention, when a query vector for searching a target image characteristic value matched with a sample image characteristic value is obtained, the query vector is segmented to obtain a plurality of query sub-vectors, and the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one by one; calculating the distance from each inquiry sub-vector in the inquiry sub-vectors to all clustering centers in the sub-space corresponding to the inquiry sub-vector, wherein the clustering centers are obtained by clustering sub-segments of a plurality of image characteristic values to be searched existing in the sub-space; acquiring an asymmetric distance between the sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all the clustering centers in the sub-space corresponding to the inquiry sub-vector; and according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the images to be searched, and acquiring the characteristic value of the target image. The query vector is divided into the query sub-vectors, the query vector with higher dimensionality is actually subdivided, the segmentation is the same as the segmentation of the characteristic value of the image to be searched into a plurality of subsections in the data storage stage, and the subdivided query sub-vectors and subsections are more beneficial to improving the similarity precision of searching; the asymmetric distance is the distance between the sample image characteristic value and the quantized image characteristic value to be searched because the distance is directly calculated, so that the calculated distance can be ensured to be closer to the real distance between the sample image characteristic value and the image characteristic value to be searched, and the searching accuracy can be improved as well; the parallel search is developed from a plurality of index objects storing less data, the efficiency is much higher than that of the independent search from one index object storing a large amount of data, and the search speed of the image characteristic value can be improved even under the condition that the data amount is continuously increased.
Referring to fig. 3, fig. 3 is a schematic flow chart of another image feature value searching method according to an embodiment of the present invention, as shown in fig. 3, including the following steps:
s21, when a query vector used for searching a target image characteristic value matched with the sample image characteristic value is obtained, segmenting the query vector to obtain a plurality of query sub-vectors, wherein the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one by one.
In the embodiment of the present invention, the step is to construct a query vector for searching the feature value of the target image in advance, so that the query vector can be directly obtained during feature value search, and then the query vector is segmented.
And S22, calculating the distance from each inquiry sub-vector in the inquiry sub-vectors to all the cluster centers in the sub-space corresponding to the inquiry sub-vector, wherein the cluster centers are obtained by clustering the subsections of the characteristic values of the images to be searched existing in the sub-spaces.
The specific implementation of step S22 is described in the above embodiments, and is also applicable to this embodiment, and will not be described in detail herein.
And S23, acquiring the asymmetric distance between the sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all the cluster centers in the sub-space corresponding to the inquiry sub-vector.
The specific implementation of step S23 is described in the above embodiments, and is also applicable to this embodiment, and will not be described in detail herein. Step S23 further includes:
s231, executing the following operations for each image feature value to be searched in the plurality of image feature values to be searched respectively: determining the clustering center of each sub-section in the plurality of sub-sections of the image characteristic value to be searched in the sub-space of the sub-section;
in the embodiment of the present invention, since the sub-segments in each subspace are clustered in the data storage stage, each sub-segment of the image feature value to be searched has a cluster center in the corresponding subspace. For example, if there are 4 sub-segments for a complete image feature value to be searched, the cluster center to which the 1-number sub-segment belongs in the 1-number subspace where the 1-number sub-segment is located is 24, the cluster center to which the 2-number sub-segment belongs in the 2-number subspace where the 2-number sub-segment is located is 8, the cluster center to which the 3-number sub-segment belongs in the 3-number subspace where the 3-number sub-segment is located is 222, and the cluster center to which the 4-number sub-segment belongs in the 4-number subspace where the 4-number sub-segment is located is 155, then it is necessary to identify the cluster center to which the 4-number sub-segments of the complete image feature value to be: number 24, number 8, number 222, and number 155.
S232, determining the distance from each query subvector in the plurality of query subvectors to the cluster center of each subsegment in the subspace of the subsegment;
and S233, performing summation operation on all the determined distances to obtain an asymmetric distance between the sample image characteristic value and the image characteristic value to be searched.
In a specific embodiment of the present invention, after determining each sub-segment in the plurality of sub-segments of each image feature value to be searched, after determining the clustering center to which the sub-segment belongs in the subspace of the sub-segment, respectively determining the distance from each query sub-vector to the clustering center to which the corresponding sub-segment belongs according to the distance from each query sub-vector to all the clustering centers in the subspace corresponding to the query sub-vector calculated in step S22, and obtaining the asymmetric distance between the sample image feature value and the image feature value to be searched by means of a summation operation. Still taking the example in step S231 for explanation, 1, 2, 3, and 4 query subvectors of one query vector respectively correspond to the subspace 1, the subspace 2, the subspace 3, and the subspace 4, so as to respectively determine the distance from the query subvector 1 to the cluster center 24 in the subspace 1, the distance from the query subvector 2 to the cluster center 8 in the subspace 2, the distance from the query subvector 3 to the cluster center 222 in the subspace 3, and the distance from the query subvector 4 to the cluster center 155 in the subspace 4, and then sum up the four distances to obtain the asymmetric distance between the sample image feature value and the image feature value to be searched.
And S24, according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the image to be searched, and acquiring the characteristic value of the target image.
Similarly, the specific implementation of step S24 is described in the above embodiments, and is also applicable to this embodiment, and will not be described in detail herein.
As an optional implementation manner, before the step of calculating the distance from each query subvector in the plurality of query subvectors to all cluster centers in the subspace corresponding to the query subvector, the method further includes:
respectively segmenting the image characteristic value to be searched aiming at each image characteristic value to be searched in the image characteristic values to be searched to obtain a plurality of subsections of the image characteristic value to be searched; wherein the plurality of subsections correspond to the plurality of subspaces one to one;
for each subspace of the plurality of subspaces, performing the following:
clustering all subsections existing in the subspace by adopting a clustering algorithm to obtain a plurality of clustering centers;
and taking the code of the clustering center as the quantization code of the sub-segments clustered under the clustering center, and storing the quantization codes of all the sub-segments in the plurality of index objects.
In the specific embodiment of the present invention, when storing a large amount of image feature values to be searched, the image feature values to be searched in a high dimension need to be segmented to obtain subspaces and subsections corresponding to the query subvectors one to one, and each subspace has a large number of subsections of the image feature values to be searched. And clustering all the subsegments in each subspace by adopting a clustering algorithm to obtain a plurality of clustering centers, and then carrying out quantitative coding on the subsegments in each subspace and storing the subsegments in the index object.
As shown in fig. 4, N128-dimensional image feature values to be searched are divided into 4 32-dimensional subsections, and 4 subspaces are obtained. And clustering in each subspace aiming at each subspace in the 4 subspaces, clustering all the subsegments in the subspaces to obtain 256 clustering centers, and then approximately representing decimal quantization codes of the subsegments clustered to the clustering centers by using the codes of the clustering centers to obtain quantization code matrixes of the 4 subsegments of all the characteristic values of the image to be searched. As shown in the right matrix of fig. 4, it can be understood that, among the N image feature values to be searched, the quantization codes of the 4 subsections of the first image feature value to be searched are 124, 56, 132, and 222, respectively. In this form, the quantized codes of all the subsections of the image feature value to be searched are stored in the index object, and since the set of quantized codes of all the subsections forms a codebook, it can be understood that each index object stores a codebook.
In the embodiment, clustering in each subspace is performed independently, each subspace starts clustering at the same time, clustering speed is greatly improved, all quantization codes are stored in a plurality of index objects instead of one index object, and data volume in each index object is relatively reduced.
As an optional implementation manner, the step of performing parallel search from a plurality of index objects used for storing the plurality of image feature values to be searched according to all the obtained asymmetric distances to obtain the target image feature value includes:
and according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the image to be searched by adopting a multi-search thread to obtain the characteristic value of the target image.
In this embodiment, the multiple search threads and the multiple index objects may be one-to-one searches, and the specific correspondence relationship is not limited herein. The multi-search thread is adopted to search in parallel from a plurality of index objects with relatively small data volume, and the searching speed is obviously faster than that of searching from one index object with large data volume.
As an optional implementation manner, the step of performing parallel search from a plurality of index objects used for storing the plurality of image feature values to be searched according to all the obtained asymmetric distances to obtain the target image feature value further includes:
selecting a preset number of target asymmetric distances from all the obtained asymmetric distances according to the sequence of the asymmetric distances from small to large;
performing parallel search from a plurality of index objects for storing the plurality of image characteristic values to be searched, and obtaining quantization codes of a plurality of subsegments of the image characteristic values to be searched corresponding to the target asymmetric distance;
and determining the characteristic value of the image to be searched corresponding to the target asymmetric distance according to the quantization codes of a plurality of subsections of the characteristic value of the image to be searched corresponding to the target asymmetric distance, and taking the characteristic value of the image to be searched corresponding to the target asymmetric distance as the characteristic value of the target image.
In an embodiment of the present invention, each image feature value to be searched corresponds to a preset identification number (ID), for example, as shown in fig. 4, the quantization codes of 4 subsections of a complete image feature value to be searched are 124, 56, 132, and 222, respectively, so that the preset identification number of the image feature value to be searched may be 1, and there are N preset identification numbers for N image feature values to be searched. According to the obtained sizes of all the asymmetric distances, selecting a preset number of target asymmetric distances with small distance values from small to large, for example: 100 or 200; and obtaining the quantization codes of a plurality of subsections of the image characteristic value to be searched corresponding to the preset number of target asymmetric distances from a plurality of index objects, determining the image characteristic value to be searched and a preset identity identification number thereof according to the obtained quantization codes of the plurality of subsections of the image characteristic value to be searched, taking all the determined image characteristic values to be searched as target image characteristic values, and returning the preset identity identification numbers thereof as search results to a searching user.
In the embodiment, the quantization codes of the multiple subsections of the image characteristic value to be searched corresponding to the preset number of target asymmetric distances are obtained from the multiple index objects, the target image characteristic value and the preset identity identification number thereof are determined according to the quantization codes of the subsections, the searching speed is improved, and the searching result with higher similarity can be returned to the user.
As an optional implementation manner, the clustering all sub-segments existing in the subspace by using a clustering algorithm to obtain a plurality of clustering centers includes:
and clustering all subsections existing in the subspace by adopting a K-means clustering algorithm to obtain a plurality of clustering centers.
In the embodiment, the clustering of the subsegments is performed by adopting a K-Means clustering algorithm, so that the clustering speed is higher, and the method is more suitable for large-scale data sets.
In this embodiment, various optional embodiments are added to the embodiment shown in fig. 1, and the search speed can be still increased even when the image feature value data amount continues to increase.
Referring to fig. 5, fig. 5 is a schematic structural diagram of an image feature value search apparatus according to an embodiment of the present invention, and as shown in fig. 5, the apparatus 5 includes:
the first segmentation module 51 is configured to, when a query vector for searching for a target image feature value matching a sample image feature value is obtained, segment the query vector to obtain a plurality of query sub-vectors, where the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one to one;
a calculating module 52, configured to calculate distances from each of the query subvectors to all cluster centers in a subspace corresponding to the query subvectors, where the cluster centers are obtained by clustering subsections of feature values of a plurality of images to be searched existing in the subspace;
an obtaining module 53, configured to obtain an asymmetric distance between the sample image feature value and each image feature value to be searched according to a distance from each query sub-vector in the multiple query sub-vectors to all cluster centers in a subspace corresponding to the query sub-vector;
and the searching module 54 is configured to perform parallel search from a plurality of index objects used for storing the plurality of image feature values to be searched according to all the acquired asymmetric distances, so as to obtain the target image feature value.
Optionally, as shown in fig. 6, the obtaining module 53 includes:
a first determining unit 531, configured to perform the following operations for each image feature value to be searched in the plurality of image feature values to be searched, respectively: determining the clustering center of each sub-section in the plurality of sub-sections of the image characteristic value to be searched in the sub-space of the sub-section;
a second determining unit 532, configured to determine a distance from each query subvector in the plurality of query subvectors to a cluster center to which each sub-segment belongs in a subspace in which the sub-segment is located;
the operation unit 533 is configured to perform summation operation on all the determined distances, so as to obtain an asymmetric distance between the sample image feature value and the image feature value to be searched.
Optionally, as shown in fig. 7, the apparatus 5 further includes:
a second segmentation module 55, configured to segment, for each to-be-searched image feature value in the to-be-searched image feature values, the to-be-searched image feature value to obtain a plurality of subsections of the to-be-searched image feature value; wherein the plurality of subsections correspond to the plurality of subspaces one to one;
a clustering module 56 configured to, for each subspace of the plurality of subspaces: clustering all subsections existing in the subspace by adopting a clustering algorithm to obtain a plurality of clustering centers;
and the storage module 57 is configured to use the codes of the cluster centers as quantization codes of sub-segments clustered under the cluster centers, and store the quantization codes of all sub-segments in the plurality of index objects.
Optionally, as shown in fig. 8, the search module 54 includes:
the searching unit 541 is configured to perform parallel search from a plurality of index objects for storing the plurality of image feature values to be searched by using a multiple search thread according to all the acquired asymmetric distances, so as to obtain the target image feature value.
Optionally, as shown in fig. 9, the search module 54 further includes:
a selecting unit 542, configured to select a preset number of target asymmetric distances from all the acquired asymmetric distances in a descending order of the asymmetric distances;
the obtaining unit 543 is configured to perform parallel search from a plurality of index objects used for storing the plurality of image feature values to be searched, and obtain quantization codes of a plurality of sub-segments of the image feature values to be searched corresponding to the target asymmetric distance;
a third determining unit 544, configured to determine, according to quantization codes of multiple subsections of the to-be-searched image feature value corresponding to the target asymmetric distance, and use the to-be-searched image feature value corresponding to the target asymmetric distance as the target image feature value.
Optionally, as shown in fig. 10, the clustering module 56 includes:
and the clustering unit 561 is configured to cluster all sub-segments existing in the subspace by using a K-means clustering algorithm to obtain a plurality of clustering centers.
Referring to fig. 11, fig. 11 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, as shown in fig. 11, including: a memory 1102, a processor 1101, and a computer program 1103 stored on the memory and executable on the processor.
The processor 1101 is configured to invoke the computer program 1103 stored in the memory 1102, and execute the following steps:
when a query vector used for searching a target image characteristic value matched with a sample image characteristic value is obtained, segmenting the query vector to obtain a plurality of query sub-vectors, wherein the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one by one;
calculating the distance from each inquiry sub-vector in the inquiry sub-vectors to all clustering centers in the sub-space corresponding to the inquiry sub-vector, wherein the clustering centers are obtained by clustering sub-segments of a plurality of image characteristic values to be searched existing in the sub-space;
acquiring an asymmetric distance between the sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all the clustering centers in the sub-space corresponding to the inquiry sub-vector;
and according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the images to be searched, and acquiring the characteristic value of the target image.
Optionally, the step of obtaining the asymmetric distance between the sample image feature value and each image feature value to be searched by the processor 1101 according to the distance from each of the plurality of query sub-vectors to all the cluster centers in the subspace corresponding to the query sub-vector includes:
respectively aiming at each image characteristic value to be searched in the plurality of image characteristic values to be searched, the following operations are carried out:
determining the clustering center of each sub-section in the plurality of sub-sections of the image characteristic value to be searched in the sub-space of the sub-section;
determining the distance from each query subvector in the plurality of query subvectors to the cluster center of each subsegment in the subspace of the subsegment;
and performing summation operation on all the determined distances to obtain the asymmetric distance between the sample image characteristic value and the image characteristic value to be searched.
Optionally, the processor 1101 is further configured to:
respectively segmenting the image characteristic value to be searched aiming at each image characteristic value to be searched in the image characteristic values to be searched to obtain a plurality of subsections of the image characteristic value to be searched; wherein the plurality of subsections correspond to the plurality of subspaces one to one;
for each subspace of the plurality of subspaces, performing the following:
clustering all subsections existing in the subspace by adopting a clustering algorithm to obtain a plurality of clustering centers;
and taking the code of the clustering center as the quantization code of the sub-segments clustered under the clustering center, and storing the quantization codes of all the sub-segments in the plurality of index objects.
Optionally, the step, executed by the processor 1101, of performing parallel search from a plurality of index objects used for storing the plurality of image feature values to be searched according to all the obtained asymmetric distances to obtain the target image feature value includes:
and according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the image to be searched by adopting a multi-search thread to obtain the characteristic value of the target image.
Optionally, the processor 1101 performs a step of performing parallel search from a plurality of index objects used for storing the plurality of image feature values to be searched according to all the obtained asymmetric distances, to obtain the target image feature value, and further includes:
selecting a preset number of target asymmetric distances from all the obtained asymmetric distances according to the sequence of the asymmetric distances from small to large;
performing parallel search from a plurality of index objects for storing the plurality of image characteristic values to be searched, and obtaining quantization codes of a plurality of subsegments of the image characteristic values to be searched corresponding to the target asymmetric distance;
and determining the characteristic value of the image to be searched corresponding to the target asymmetric distance according to the quantization codes of a plurality of subsections of the characteristic value of the image to be searched corresponding to the target asymmetric distance, and taking the characteristic value of the image to be searched corresponding to the target asymmetric distance as the characteristic value of the target image.
Optionally, the step of clustering all sub-segments existing in the subspace by using the clustering algorithm executed by the processor 1101 to obtain a plurality of clustering centers includes:
and clustering all subsections existing in the subspace by adopting a K-means clustering algorithm to obtain a plurality of clustering centers.
That is, in the embodiment of the present invention, the steps in the image feature value search method described above are implemented when the processor 1101 of the electronic device 11 executes the computer program 1103, so that the search speed of the image feature value can be increased even when the data amount continues to increase.
The electronic device 11 may be, for example, a mobile phone, a computer, a notebook computer, a tablet computer, a palm computer, a wearable device, and the like. The electronic device 11 may include, but is not limited to, a processor 1101, a memory 1102. Those skilled in the art will appreciate that the schematic diagram is merely an example of the electronic device 11 and does not constitute a limitation of the electronic device 11 and may include more or less components than those shown, or some components in combination, or different components.
It should be noted that, since the steps in the image feature value searching method described above are implemented when the processor 1101 of the electronic device 11 executes the computer program 1103, all embodiments of the image feature value searching method described above are applicable to the electronic device 11, and the same or similar beneficial effects can be achieved.
The embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored, and the computer program, when executed by a processor, implements the steps in the image feature value searching method described above.
That is, in a specific embodiment of the present invention, a computer program of a computer-readable storage medium, when executed by a processor, implements the steps in the above-described image feature value search method, thereby being able to improve the search speed of image feature values even if the amount of data continues to increase.
Illustratively, the computer program of the computer-readable storage medium comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, and the like. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution medium, and the like.
It should be noted that, since the computer program of the computer-readable storage medium is executed by the processor to implement the steps in the image feature value searching method, all the embodiments of the image feature value searching method are applicable to the computer-readable storage medium, and the same or similar beneficial effects can be achieved.
The above embodiments of the present invention are described in detail, and the principle and the implementation of the present invention are explained by applying specific embodiments, and the above description of the embodiments is only used to help understanding the method of the present invention and the core idea thereof; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (10)
1. A method for searching for an image feature value, the method comprising:
when a query vector used for searching a target image characteristic value matched with a sample image characteristic value is obtained, segmenting the query vector to obtain a plurality of query sub-vectors, wherein the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one by one;
calculating the distance from each inquiry sub-vector in the inquiry sub-vectors to all clustering centers in the sub-space corresponding to the inquiry sub-vector, wherein the clustering centers are obtained by clustering sub-segments of a plurality of image characteristic values to be searched existing in the sub-space;
acquiring an asymmetric distance between the sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all the clustering centers in the sub-space corresponding to the inquiry sub-vector;
and according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the images to be searched, and acquiring the characteristic value of the target image.
2. The method of claim 1, wherein the step of obtaining the asymmetric distance between the sample image feature value and each image feature value to be searched according to the distance from each query subvector in the plurality of query subvectors to all cluster centers in the subspace corresponding to the query subvector comprises:
respectively aiming at each image characteristic value to be searched in the plurality of image characteristic values to be searched, the following operations are carried out:
determining the clustering center of each sub-section in the plurality of sub-sections of the image characteristic value to be searched in the sub-space of the sub-section;
determining the distance from each query subvector in the plurality of query subvectors to the cluster center of each subsegment in the subspace of the subsegment;
and performing summation operation on all the determined distances to obtain the asymmetric distance between the sample image characteristic value and the image characteristic value to be searched.
3. The method of claim 1, wherein prior to the step of calculating the distance of each query subvector in the plurality of query subvectors to all cluster centers in the subspace corresponding to the query subvector, the method further comprises:
respectively segmenting the image characteristic value to be searched aiming at each image characteristic value to be searched in the image characteristic values to be searched to obtain a plurality of subsections of the image characteristic value to be searched; wherein the plurality of subsections correspond to the plurality of subspaces one to one;
for each subspace of the plurality of subspaces, performing the following:
clustering all subsections existing in the subspace by adopting a clustering algorithm to obtain a plurality of clustering centers;
and taking the code of the clustering center as the quantization code of the sub-segments clustered under the clustering center, and storing the quantization codes of all the sub-segments in the plurality of index objects.
4. The method according to claim 1, wherein the step of performing parallel search from a plurality of index objects for storing the plurality of image feature values to be searched according to all the obtained asymmetric distances to obtain the target image feature value comprises:
and according to all the acquired asymmetric distances, performing parallel search from a plurality of index objects for storing the characteristic values of the image to be searched by adopting a multi-search thread to obtain the characteristic value of the target image.
5. The method according to claim 3, wherein the step of performing parallel search from a plurality of index objects for storing the plurality of image feature values to be searched according to all the obtained asymmetric distances to obtain the target image feature value further comprises:
selecting a preset number of target asymmetric distances from all the obtained asymmetric distances according to the sequence of the asymmetric distances from small to large;
performing parallel search from a plurality of index objects for storing the plurality of image characteristic values to be searched, and obtaining quantization codes of a plurality of subsegments of the image characteristic values to be searched corresponding to the target asymmetric distance;
and determining the characteristic value of the image to be searched corresponding to the target asymmetric distance according to the quantization codes of a plurality of subsections of the characteristic value of the image to be searched corresponding to the target asymmetric distance, and taking the characteristic value of the image to be searched corresponding to the target asymmetric distance as the characteristic value of the target image.
6. The method according to claim 3, wherein the step of clustering all sub-segments existing in the subspace using a clustering algorithm to obtain a plurality of clustering centers comprises:
and clustering all subsections existing in the subspace by adopting a K-means clustering algorithm to obtain a plurality of clustering centers.
7. An apparatus for searching for an image feature value, comprising:
the system comprises a first segmentation module, a second segmentation module and a third segmentation module, wherein the first segmentation module is used for segmenting a query vector to obtain a plurality of query sub-vectors when the query vector used for searching a target image characteristic value matched with a sample image characteristic value is obtained, and the plurality of query sub-vectors correspond to a plurality of pre-obtained subspaces one by one;
the computing module is used for computing the distance from each query subvector in the plurality of query subvectors to all the clustering centers in the subspace corresponding to the query subvectors, and the clustering centers are obtained by clustering subsections of a plurality of image characteristic values to be searched existing in the subspaces;
the acquisition module is used for acquiring the asymmetric distance between the sample image characteristic value and each image characteristic value to be searched according to the distance from each inquiry sub-vector in the inquiry sub-vectors to all the clustering centers in the sub-space corresponding to the inquiry sub-vector;
and the searching module is used for performing parallel searching from a plurality of index objects for storing the plurality of image characteristic values to be searched according to all the acquired asymmetric distances to obtain the target image characteristic value.
8. The apparatus of claim 7, wherein the obtaining module comprises:
a first determining unit, configured to perform the following operations for each image feature value to be searched in the plurality of image feature values to be searched, respectively: determining the clustering center of each sub-section in the plurality of sub-sections of the image characteristic value to be searched in the sub-space of the sub-section;
a second determining unit, configured to determine a distance from each query subvector in the multiple query subvectors to a cluster center to which each sub-segment belongs in a subspace in which the sub-segment is located;
and the operation unit is used for carrying out summation operation on all the determined distances to obtain the asymmetric distance between the sample image characteristic value and the image characteristic value to be searched.
9. An electronic device, comprising: memory, processor and computer program stored on the memory and executable on the processor, the processor implementing the steps in the image feature value search method according to any one of claims 1 to 6 when executing the computer program.
10. A computer-readable storage medium, characterized in that a computer program is stored thereon, which computer program, when being executed by a processor, carries out the steps in the image feature value search method according to any one of claims 1 to 6.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811341714.3A CN111177438B (en) | 2018-11-12 | 2018-11-12 | Image characteristic value searching method and device, electronic equipment and storage medium |
| PCT/CN2018/124832 WO2020098110A1 (en) | 2018-11-12 | 2018-12-28 | Image feature value searching method and apparatus, electronic device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811341714.3A CN111177438B (en) | 2018-11-12 | 2018-11-12 | Image characteristic value searching method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111177438A true CN111177438A (en) | 2020-05-19 |
| CN111177438B CN111177438B (en) | 2023-05-12 |
Family
ID=70646050
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811341714.3A Active CN111177438B (en) | 2018-11-12 | 2018-11-12 | Image characteristic value searching method and device, electronic equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111177438B (en) |
| WO (1) | WO2020098110A1 (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111797259A (en) * | 2020-07-10 | 2020-10-20 | 杭州慕锐科技有限公司 | Rapid image searching method for wrinkled deformation |
| CN114817717A (en) * | 2022-04-21 | 2022-07-29 | 国科华盾(北京)科技有限公司 | Search method, search device, computer equipment and storage medium |
| WO2023065697A1 (en) * | 2021-10-21 | 2023-04-27 | 深圳云天励飞技术股份有限公司 | Product quantization search method and apparatus, and terminal and storage medium |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112307035B (en) * | 2020-11-26 | 2024-01-05 | 深圳云天励飞技术股份有限公司 | Method and device for managing characteristic value ID, electronic equipment and storage medium |
| CN113065011B (en) * | 2021-03-17 | 2024-01-16 | 北京沃东天骏信息技术有限公司 | Picture determination method and device |
| CN116796021B (en) * | 2023-08-28 | 2023-12-05 | 上海任意门科技有限公司 | Image retrieval method, system, electronic device and medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105117407A (en) * | 2015-07-27 | 2015-12-02 | 电子科技大学 | Image retrieval method for cluster-based distance direction histogram |
| CN106326288A (en) * | 2015-06-30 | 2017-01-11 | 阿里巴巴集团控股有限公司 | Image search method and apparatus |
| CN106886553A (en) * | 2016-12-27 | 2017-06-23 | 浙江宇视科技有限公司 | A kind of image search method and server |
| US20170262478A1 (en) * | 2014-09-09 | 2017-09-14 | Thomson Licensing | Method and apparatus for image retrieval with feature learning |
| CN107766492A (en) * | 2017-10-18 | 2018-03-06 | 北京京东尚科信息技术有限公司 | A kind of method and apparatus of picture search |
| CN108241745A (en) * | 2018-01-08 | 2018-07-03 | 阿里巴巴集团控股有限公司 | The processing method and processing device of sample set, the querying method of sample and device |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103336795B (en) * | 2013-06-09 | 2017-03-08 | 华中科技大学 | Video index method based on multiple features |
| CN108733780B (en) * | 2018-05-07 | 2020-06-23 | 浙江大华技术股份有限公司 | Picture searching method and device |
-
2018
- 2018-11-12 CN CN201811341714.3A patent/CN111177438B/en active Active
- 2018-12-28 WO PCT/CN2018/124832 patent/WO2020098110A1/en active Application Filing
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170262478A1 (en) * | 2014-09-09 | 2017-09-14 | Thomson Licensing | Method and apparatus for image retrieval with feature learning |
| CN106326288A (en) * | 2015-06-30 | 2017-01-11 | 阿里巴巴集团控股有限公司 | Image search method and apparatus |
| CN105117407A (en) * | 2015-07-27 | 2015-12-02 | 电子科技大学 | Image retrieval method for cluster-based distance direction histogram |
| CN106886553A (en) * | 2016-12-27 | 2017-06-23 | 浙江宇视科技有限公司 | A kind of image search method and server |
| CN107766492A (en) * | 2017-10-18 | 2018-03-06 | 北京京东尚科信息技术有限公司 | A kind of method and apparatus of picture search |
| CN108241745A (en) * | 2018-01-08 | 2018-07-03 | 阿里巴巴集团控股有限公司 | The processing method and processing device of sample set, the querying method of sample and device |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111797259A (en) * | 2020-07-10 | 2020-10-20 | 杭州慕锐科技有限公司 | Rapid image searching method for wrinkled deformation |
| WO2023065697A1 (en) * | 2021-10-21 | 2023-04-27 | 深圳云天励飞技术股份有限公司 | Product quantization search method and apparatus, and terminal and storage medium |
| CN114817717A (en) * | 2022-04-21 | 2022-07-29 | 国科华盾(北京)科技有限公司 | Search method, search device, computer equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111177438B (en) | 2023-05-12 |
| WO2020098110A1 (en) | 2020-05-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111177438B (en) | Image characteristic value searching method and device, electronic equipment and storage medium | |
| Celik et al. | Content based image retrieval with sparse representations and local feature descriptors: A comparative study | |
| KR102449191B1 (en) | Method and apparatus for supporting machine learning algorithms and data pattern matching in ethernet solid state drive | |
| Shen et al. | Inductive hashing on manifolds | |
| Weiss et al. | Spectral hashing | |
| Lin et al. | A general two-step approach to learning-based hashing | |
| Wu et al. | Multiscale quantization for fast similarity search | |
| Gong et al. | Compressing deep convolutional networks using vector quantization | |
| CN105912611B (en) | A kind of fast image retrieval method based on CNN | |
| Lin et al. | Supervised hashing using graph cuts and boosted decision trees | |
| Wei et al. | Projected residual vector quantization for ANN search | |
| WO2016037844A1 (en) | Method and apparatus for image retrieval with feature learning | |
| Chiu et al. | Learning to index for nearest neighbor search | |
| US20230123941A1 (en) | Multiscale Quantization for Fast Similarity Search | |
| CN114238329A (en) | Vector similarity calculation method, device, equipment and storage medium | |
| CN113656373A (en) | Method, device, equipment and storage medium for constructing retrieval database | |
| Amato et al. | Aggregating binary local descriptors for image retrieval | |
| Valem et al. | Unsupervised similarity learning through Cartesian product of ranking references | |
| Yu et al. | Bilinear optimized product quantization for scalable visual content analysis | |
| Mu et al. | Accelerated large scale optimization by concomitant hashing | |
| Avrithis | Quantize and Conquer: A dimensionality-recursive solution to clustering, vector quantization, and image retrieval | |
| Dammak et al. | Histogram of dense subgraphs for image representation | |
| EP3192010A1 (en) | Image recognition using descriptor pruning | |
| Sun et al. | Automating nearest neighbor search configuration with constrained optimization | |
| CN110209895B (en) | Vector retrieval method, device and equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |