Detailed Description
Preferred embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While the preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
The term "include" and variations thereof as used herein is meant to be inclusive in an open-ended manner, i.e., "including but not limited to". Unless specifically stated otherwise, the term "or" means "and/or". The term "based on" means "based at least in part on". The terms "one example embodiment" and "one embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first," "second," and the like may refer to different or the same object.
As described above, in the conventional image retrieval method based on text description, since manual intervention is required when an image is labeled, it is not suitable for retrieving massive image data. However, the conventional image retrieval method based on image content needs to calculate and store the feature vectors at high latitude, so that a large amount of calculation resources and time need to be consumed, and the requirement of the user on query real-time performance cannot be quickly responded.
To address, at least in part, one or more of the above problems, as well as other potential problems, example embodiments of the present disclosure propose a solution for image retrieval. The scheme comprises the following steps: extracting the features of the image to be retrieved through a neural network model to generate the feature vector of the image to be retrieved, wherein the neural network model is trained through multiple samples; performing locality sensitive hash calculation on the feature vector of the image to be retrieved to generate a first hash value and a second hash value, wherein the first hash value is used for identifying image grouping information of the image to be retrieved, and the second hash value is used for representing image feature information of the image to be retrieved; querying a first predetermined image database based on the first hash value to obtain image feature information of a plurality of candidate object images, image grouping information of the plurality of candidate object images being matched with the first hash value; calculating similarities of the second hash values and image feature information of the plurality of candidate object images, respectively, so as to sort the plurality of candidate object images based on the similarity calculation results, the image grouping information and the image feature information of the plurality of candidate object images being generated via locality sensitive hash calculation; and determining an image retrieval result matched with the image to be retrieved based on the similarity calculation result and a preset similarity threshold.
In the above-described aspect, the present disclosure can query image feature information of a plurality of candidate object images whose image grouping information matches based on a first hash value of a binary image to be retrieved generated via locality sensitive hashing without performing query and similarity calculation based on an image feature vector of a high latitude, and then determine an image retrieval result based on a similarity calculation result of a second hash value of the image to be retrieved and the image feature information of the candidate object images. Compared with the traditional image retrieval method based on the content, the image retrieval method based on the locality sensitive hashing calculation can map the original image feature vector into the first hashing value and the second hashing value which are relatively compact in hash mode, so that resources consumed by calculation and storage can be greatly reduced, and time required by query response is shortened. Therefore, the method and the device can effectively reduce the computing and storage resources required by massive image retrieval and shorten the time required by query response.
Fig. 1 shows a schematic diagram of a system 100 for implementing a method for image retrieval according to an embodiment of the present disclosure. As shown in fig. 1, the system 100 includes: a data acquisition unit 112, a feature extraction unit 114, a hash value generation unit 116, an image grouping information search unit 120, a candidate object image feature information acquisition unit 122, a similarity calculation and ranking unit 124, an image retrieval result determination unit 126, an image retrieval result output unit 128, an object image database 160. In some embodiments, the system 100 further comprises: server 140, network 150.
In some embodiments, the data acquisition unit 112, the feature extraction unit 114, the hash value generation unit 116, the image grouping information query unit 120, the candidate object image feature information acquisition unit 122, the similarity calculation and ranking unit 124, the image retrieval result determination unit 126, and the image retrieval result output unit 128 may be configured on one or more computing devices 130. Computing device 130 may interact with object image database 160 and computing device 130 may interact with server 140 via wired or wireless means (e.g., network 150).
With respect to the computing device 130, it is used for managing the object images and the object image database, and for retrieving the object image database based on the image to be retrieved so as to determine the image retrieval result matching the image to be retrieved. The object image is, for example and without limitation, a commodity image. The object is, for example and without limitation, a commodity included in the commodity image.
With regard to the management of the object image and the object image database, the computing device 130 is specifically configured to perform hash computation on the feature vector with the obtained object image to generate image grouping information and image feature information in the form of binary hash values; splicing the image grouping information, the object category and the object attribute identification so as to generate an identification of a hash bucket; establishing a first mapping relation between the identifier of the hash bucket and the identifier of the one or more object images; establishing a second mapping relation between the object image identification and the image characteristic information of the object image, and storing the first mapping relation and the second mapping relation in a first preset object image database and a second preset object image database of the object image database. Object categories are for example and not limited to categories of goods; the object attribute identifies, for example and without limitation, brand information for the item. The object image identification is, for example, an object identification (such as a commodity identification).
Regarding retrieving the matched image retrieval result based on the image to be retrieved, the computing device 130 is specifically configured to generate a feature vector of the image to be retrieved, and perform locality sensitive hash calculation on the feature vector of the image to be retrieved to generate a first hash value and a second hash value; inquiring a plurality of candidate object images of which the image grouping information is matched with the first hash value in a first preset object image database; and respectively calculating the similarity of the second hash value and the image characteristic information of the candidate object images, and determining an image retrieval result matched with the image to be retrieved based on the similarity calculation result and a preset similarity threshold value. In some embodiments, computing device 130 may have one or more processing units, including special purpose processing units such as GPUs, FPGAs, ASICs, and general purpose processing units such as CPUs. In addition, one or more virtual machines may also be running on each computing device.
As for the data acquisition unit 112, it is used to acquire an image to be retrieved and all object images (for example, commodity images). For example, the data acquisition unit 112 may obtain the image to be retrieved 160 and all object images 162 from the server 140 or input by the user from an input device of the computing device 130 via the network 150. The all-object image 162 is, for example, an image of all approved merchandise on the internet e-commerce platform. In some embodiments, the data obtaining unit 112 is further configured to obtain an instruction about the same image retrieval or an instruction about similar image retrieval, which is input by a user.
A feature extraction unit 114, configured to extract features of the image to be retrieved via a multi-sample trained neural network model to generate a feature vector associated with the image to be retrieved; and respectively extracting the features of all the object images through the same neural network model to generate the feature vectors of all the object images.
The hash value generation unit 116 performs locality sensitive hash calculation on the feature vector of each object image to generate image packet information (or "packet hash value") and image feature information (or "feature hash value"), and performs locality sensitive hash calculation on the image to be retrieved to generate a first hash value and a second hash value.
The object image database 160 stores association information of object images (for example, commodity images). The object image database 160 includes, for example: a first predetermined image database and a second predetermined image database. The first predetermined image database is configured to store at least: a first mapping relationship between the identity of the hash bucket and the identity of the one or more objects. The second predetermined image database is configured to store at least: and a second mapping relationship between the object identifier and the image characteristic information of the object image. Multiple object image identifications of image packet information having the same binary hash value are stored in the same "hash bucket".
The image grouping information query unit 120 is configured to query a first predetermined image database in the object image database 160 based on the first hash value generated by the hash value generation unit 116, so as to obtain image feature information of a plurality of candidate object images whose image grouping information matches the first hash value.
And a similarity calculation and sorting unit 124 for calculating the similarities of the second hash value and the image feature information of the plurality of candidate object images determined by the image grouping information inquiring unit 120, respectively, so as to sort the plurality of candidate object images based on the similarity calculation result.
Regarding the image retrieval result determining unit 126, it is used to determine the image retrieval result matching the image to be retrieved based on the similarity calculation result output by the similarity calculation and sorting unit 124 and a predetermined similarity threshold.
And an image retrieval result output unit 128 for outputting an image retrieval result matched with the image to be retrieved to the user.
A method 200 for image retrieval according to an embodiment of the present disclosure will be described below in conjunction with fig. 2. Fig. 2 shows a flow diagram of a method 200 for image retrieval according to an embodiment of the present disclosure. It should be understood that the method 200 may be performed, for example, at the electronic device 700 depicted in fig. 7. May also be executed at the computing device 130 depicted in fig. 1. It should be understood that method 200 may also include additional acts not shown and/or may omit acts shown, as the scope of the disclosure is not limited in this respect.
At block 202, the computing device 130 extracts features of the image to be retrieved via a neural network model via training of multiple samples to generate feature vectors for the image to be retrieved.
As for the neural network model, it is, for example, a network model using an algorithm based on Xception, VGG16, ResNet50V2, resenext 50, inclusion V3, MobileNet, or densnet 121. Compared with the image feature extraction algorithm which is usually adopted by the traditional image retrieval method, the feature vector which is associated with the image to be retrieved is generated through the neural network model, so that richer features of the image to be retrieved can be extracted, and the accuracy of image retrieval is improved. Fig. 3 schematically shows a schematic diagram of a neural network model according to the present disclosure. As shown in fig. 3, 310 indicates an input image of 3 channels, for example, 224 x 224 in size. The input image may be an image to be retrieved acquired by the computing device 130 at the time of retrieval, or may be an object image (for example, a commodity image) acquired by the computing device 130 at the time of configuring an object image database. 312 represents two layers of convolution and activation layers (Conv + Relu), each layer of convolution and activation layers 312 including a convolution operation and an activation layer following the convolution operation. The input size of the convolution and activation layer 312 becomes an image of 64 channels of 224 x 224. 314 represents the maximum pooling layer (max pooling); 316 represents the convolution and activation layer of two layers (Conv + Relu). The input size of the max-pooling layer 314 and the mesh layer 316 becomes an image of 128 channels, which is 112 x 112. 318 represents the max pooling layer (max pooling), 320 represents the convolution and activation layer of three layers (Conv + Relu), and the input size of the max pooling layer 318, convolution and activation layer 320 becomes 256 images of 56 x 56. 28 x 28 images of 512 channels. 322 represents the max pooling layer (max pooling), 324 represents the convolution and activation layer of three layers (Conv + Relu), and the input size of the max pooling layer 322, convolution and activation layer 324 becomes an image of 512 channels of 28 × 28. The input size of the max pooling layer (max pooling) 326 and the convolution of the three layers and activation layer (Conv + Relu) 328 becomes an image of 14 × 14 channels 512. 330 represents the convolution and activation layer (Conv + Relu) which inputs an image of 512 channels of size 7 x 7. 332 represents the fully connected and active layer (full connected + Relu), whose input size becomes, for example, an image of 4096 channels of 1 x 1. 334 denotes softmax processing. The feature vector of the image to be retrieved generated at block 202 is high in dimension, for example up to thousands of dimensions, as shown in fig. 3.
At block 204, the computing device 130 performs a locality sensitive hash calculation on the feature vector of the image to be retrieved to generate a first hash value and a second hash value, the first hash value being used to identify image grouping information of the image to be retrieved, and the second hash value being used to characterize image feature information of the image to be retrieved. By employing the locality sensitive hash computation described above, high-dimensional (e.g., up to thousands of dimensions) feature vectors generated via neural network models are significantly reduced in dimension, e.g., to relatively low-dimensional binary hash values.
In some embodiments, the computing device 130 may employ a Locality-Sensitive Hashing (LSH) to hash the feature vectors of the image to be retrieved. The advantage of using the locality sensitive hashing algorithm is that: the merging of similar pictures can be completed without training, and the defects of the traditional algorithms such as clustering and the like that the similar pictures need to be fitted again along with the expansion of a data set are avoided.
In some embodiments, the computing device 130 may perform a hash calculation on the feature vector of the image to be retrieved using a p-Stable (p-Stable) distributed LSH algorithm to generate a first hash value and a second hash value. The above-mentioned p stability profile can be described as: for a distribution D over a real number set R, if P is present>=0, for any n real numbers v1, …, vn and n variables X1, …, Xn satisfying D distribution, random variables Σ iviXi (vector dot product, one number) and (Σ i | vi | yp)1/pX(P [ step of]Norm, a number) has the same distribution, where X is a random variable that follows the distribution of D, then D is called a p-stable distribution. For any p e (0, 2)]There is a stable distribution: p =1 is the Cauchy distribution with a probability density function of c (x) =1/[ pi (1+ x2)](ii) a p =2 is a gaussian distribution and the probability density function is g (x) =1/(2 pi) 1/2 × e-x ^ 2/2.
The feature vectors of the image are mapped to hash values by using an LSH algorithm based on a p-stable distribution. The hash function is locally sensitive, so if the feature vector v1 of the image to be retrieved and the feature vector v2 of the candidate object image are close to each other, the hash values mapped by the hash function will be the same, and the probability of being hashed into the same hash bucket will be high.
The algorithm of the locality sensitive hash calculation described above is described below with reference to equation (1).
hash_vector = A * v (1)
In the above equation (1), the hash _ vector represents a hash vector (or referred to as "hash value") calculated via a locality sensitive hash. A represents a hash function mapping matrix. v represents a feature vector of an image generated via features extracted by the neural network model. v is, for example, a feature vector of an image to be retrieved at the time of retrieval, or a feature vector of an object image at the time of arranging an object image database. And (3) judging the elements in the characteristic vector v through a hash function mapping matrix A shown in a formula (1), wherein if the elements are larger than 0, the corresponding characteristic of the hash vector hash _ vector is 1, and otherwise, the corresponding characteristic is 0.
An example of the hash function mapping matrix a is described below in connection with equation (2), for example.
A=Shape:(N,D) (2)
In the above formula (2), N represents the mapping dimension of the hash function mapping function, i.e., the dimension of the hash vector (or "hash value") hash _ vector. D represents the dimension of the feature vector v of the image. Shape: () For example representing a locality sensitive hash function based on a stable distribution of p. Based on the p-stable distribution, the hash-mapping distances A v1-A v2 and | v1-v2| of the feature vectors v1 and v2 of the two imagespThe distribution of X is the same. Therefore, the high-dimensional feature vector of the image can be effectively approximated by utilizing the locally sensitive Hash calculation of the p stable distribution, and the dimension of the high-dimensional feature vector of the image is reduced while the measurement distance is ensured. For example, reducing the D-dimensional feature vector v of the image to a hash value for characterizing the image feature of N-dimension (e.g., 256-dimension or 512-dimension), such as a second hash value hash _ vector2 or image feature information, wherein the second hash value is used for characterizing the image feature of the image to be retrieved; the image feature information is used to characterize image features of an object image (e.g., a commodity image).
As to the manner of generating the first hash value and the second hash value, it includes, for example: based on a first preset Hash mapping dimension, carrying out partial sensitive Hash calculation based on P stable distribution on the characteristic vector of the image to be retrieved so as to generate a first Hash value; and performing partial sensitive hash calculation based on P stable distribution on the characteristic vector of the image to be retrieved based on a second preset hash mapping dimension to generate a second hash value, wherein the second preset hash mapping dimension is larger than the first preset hash mapping dimension.
With respect to the first predetermined Hash mapping dimension, it is determined, for example, based on a Hash bucket (Hash Buckets) dimension. The size of the Hash Buckets (Hash Buckets) dimension is related to the number of image features stored in each Hash bucket. For example, the first predetermined mapping dimension is configured to be 50, i.e., the dimension of the first hash value hash _ vector1 (i.e., packet hash) is configured to be 50, at which time the number of image features per hash bucket is around 30. By adopting the configuration, the image difference of the hash bucket is not large, and the excessive calculation amount in the image retrieval can be avoided.
With respect to the predetermined hash mapping dimension, it is determined based on the image feature hash dimension, for example. The second predetermined hash-mapping dimension is generally larger than the predetermined hash-mapping dimension. By configuring the second predetermined hash mapping dimension to be a high dimension, it is beneficial to reduce distortion of the characterized image features, thereby improving the accuracy of the retrieval. For example, the second predetermined hash mapping dimension is configured to be 256 or 512, in particular 512. For example, by configuring the second hash value hash _ vector2 or the dimension of the image feature information to be 512, the number of images in the hash bucket is only 1, which indicates that the hash value of 512 dimensions at this time can be used to characterize the image feature. In the above solution, by configuring the first predetermined hash mapping dimension as 50 and the second predetermined hash mapping dimension as 512, the present disclosure can not only further achieve that the difference of the images in the hash bucket is not large, so as to avoid an excessive calculation amount at the time of image retrieval, but also further reduce distortion of the characterized image features to guarantee accuracy of the retrieval result while improving the retrieval calculation and response time.
At block 206, the computing device 130 queries a first predetermined image database based on the first hash value to obtain image feature information for a plurality of candidate object images, image grouping information for the plurality of candidate object images matching the first hash value.
Regarding the method for obtaining multiple candidate object images, in some embodiments, after the image to be retrieved is mapped by the LSH hash function at block 206, two binary hash values, i.e. a first hash value and a second hash value, are generated, the "hash bucket" in which the object image identifiers having the same grouping information (image index) are located is queried through the first hash value, and then the image feature information associated with all the candidate object image identifiers stored in the "hash bucket" is obtained for the similarity calculation at the subsequent block 208. The method of obtaining a plurality of candidate images will be further described below in conjunction with fig. 6. Here, the description is omitted.
At block 208, the computing device 130 calculates similarities of the second hash values and the image feature information of the plurality of candidate object images, respectively, to sort the plurality of candidate object images based on the similarity calculation results, the image grouping information and the image feature information of the plurality of candidate object images being generated via the locality sensitive hash calculation.
As for a method of calculating the similarity between the second hash value and the image feature information, for example, it is: the calculation device 130 calculates the similarity of the second hash value to the image feature information of each of the plurality of candidate object images, respectively, based on the hamming distance.
Regarding the calculation of the hamming distance between the second hash value hash _ vector2 and the image feature information (i.e., feature hash value) of each candidate image, it is, for example: the two binary hash vectors of the second hash value hash _ vector2 and the image feature information are subjected to exclusive or operation, and the number of 1 is counted, which indicates the number of different corresponding bits of the two (same length) hash vectors (i.e., hash _ vector2 and the image feature information). By adopting the Hamming distance, the XOR exclusive operation of the internal arithmetic unit of the computer can be used for carrying out similarity calculation, so that the calculation of the Hamming distance can be completed within microsecond order, and the time required by single image retrieval query response can be further shortened.
At block 210, the computing device 130 determines an image retrieval result that matches the image to be retrieved based on the similarity calculation result and a predetermined similarity threshold.
Regarding the predetermined similarity threshold, it may be preset based on the application scene of the retrieval image of the present disclosure, or may be determined according to user input, for example, when the user inputs the image to be retrieved, an instruction regarding the same image retrieval or similar image retrieval is input, and the corresponding predetermined similarity threshold is determined to match the instruction input by the user. The method 500 for determining the predetermined similarity threshold will be further defined with reference to fig. 5, and will not be described herein again.
In the scheme, the original image feature vector is subjected to Hash mapping to form the first Hash value and the second Hash value which are relatively compact through the image retrieval method based on the locality sensitive Hash calculation, so that resources consumed by calculation and storage can be greatly reduced, and the time required by query response is shortened. Therefore, the method and the device can effectively reduce the computing and storage resources required by massive image retrieval and shorten the time required by query response.
A method 400 for managing object image data according to an embodiment of the present disclosure will be described below in conjunction with fig. 4. FIG. 4 shows a flow diagram of a method 400 for managing object image data in accordance with an embodiment of the present disclosure. It should be understood that method 400 may be performed, for example, at electronic device 700 depicted in fig. 7. May also be executed at the computing device 130 depicted in fig. 1. It should be understood that method 400 may also include additional acts not shown and/or may omit acts shown, as the scope of the disclosure is not limited in this respect.
At block 402, the computing device 130 obtains a plurality of object images, the object images associated with object categories and object attribute identifications, the objects included in the object images. The object is, for example, an article of merchandise, such as a sneaker. The object image is, for example, a commodity image such as a sneaker image. The object category is, for example, a commodity category, such as basketball shoes. The object attribute identification is, for example, a brand of merchandise, such as a merchant brand of basketball shoes.
At block 404, the computing device 130 performs a locality sensitive hash calculation on the feature vectors of the object image generated via the neural network model to generate image grouping information and image feature information for the object image. The locality sensitive hash calculation is the same as the locality sensitive hash calculation at block 204, which is, for example, a LSH calculation based on a p-stable distribution.
In some embodiments, when configuring an object image database, each object image is mapped into two binary hash values, i.e. image grouping information and image feature information, e.g. respectively, via the LSH algorithm: image grouping information in the form of a 50-dimensional binary hash value, and image feature information in the form of a 512-dimensional binary hash value.
At block 406, the computing device 130 concatenates the image grouping information, the object class, and the object attribute identification to generate an identification of the hash bucket. The object image identifications of the image packet information having the same binary hash value are stored in the same "hash bucket".
The object image identification is for example an identification of an object comprised by the object image. By enabling the identifier of the hash bucket to comprise the image grouping information, the object category and the object attribute identifier of the object image, the identifier of the hash bucket indicates the information about the object included in the image and the category and attribute identifier of the object, and the accuracy of image retrieval is improved. For example, when the category and attribute identification of the object to be checked and the first hash value of the image to be retrieved are matched with the information indicated by the identification of the hash bucket, the image feature information of a plurality of candidate object images associated with the identification of the hash bucket is obtained for similarity calculation with the second hash value.
At block 408, the computing device 130 establishes a first mapping between the identity of the hash bucket and the identity of the one or more object images to store the first mapping in a first predetermined image database.
At block 410, the computing device 130 establishes a second mapping relationship between the identification of the subject image and the image characteristic information of the subject image to store the second mapping relationship in a second predetermined image database. The first predetermined image database and the second predetermined image database are used for respectively storing a first mapping relation between the identifier of the hash bucket and the identifier of the object image and a second mapping relation between the identifier of the object image and the image feature information of the object image, so that image feature indexing and organization management are facilitated to be clearer.
In a conventional image retrieval method, for example, a storage manner of a feature vector of a hash value-related image is adopted. Since the feature vectors of the image have high dimensionality, the feature vectors of the image are directly stored, which requires a large amount of storage space, for example, about 80G of storage space is required for 200 ten thousand object images. In addition, the hash algorithm needs multiple sets of hashes to maintain accuracy, and therefore, the feature vectors of the image are further stored multiple times, which requires a larger storage space. Therefore, the requirement on storage space is high, the image query performance is low, and the retrieval and quick response of massive image data are difficult.
The method comprises the steps of respectively representing grouping information and image characteristics of each object image through two binary hash values, and configuring a storage and index mode of image data by establishing a mapping relation between a hash bucket identifier and an object image identifier and a mapping relation between the object image identifier and image characteristic information of the object image. It is possible to reduce the storage space of the object image data and improve the retrieval response time for the object image data. For example, 900 ten thousand object images, only 64G of storage space is required, and the query response time can be controlled around 100ms by load balancing.
Fig. 5 shows a flow diagram of a method 500 for determining a predetermined similarity threshold according to an embodiment of the present disclosure. It should be understood that method 500 may be performed, for example, at electronic device 700 depicted in fig. 7. May also be executed at the computing device 130 depicted in fig. 1. It should be understood that method 500 may also include additional acts not shown and/or may omit acts shown, as the scope of the disclosure is not limited in this respect.
At block 502, the computing device 130 determines whether an instruction for the same image retrieval is detected.
If the computing device 130 determines that an instruction is detected for the same image retrieval, the predetermined similarity threshold is determined to be a first threshold at block 504.
If the computing device 130 determines that no instructions are detected for the same image retrieval, a determination is made at block 506 whether instructions are detected for a proximate image retrieval.
If the computing device 130 determines that an instruction is detected for proximate image retrieval, at block 508, the computing device 130 determines that the predetermined similarity threshold is a second threshold, the second threshold being less than the first threshold.
By adopting the means, the image retrieval result can be automatically matched with the differential retrieval requirements of the same image retrieval and the similar image retrieval.
Instructions for same image retrieval, instructions for similar image retrieval, which may be input by a user directly via computing device 130 or input to computing device 130 by server 130 and network 150; it may also be determined automatically based on different retrieval application scenarios, for example, if the application scenario is an accurate image retrieval, the predetermined similarity threshold needs to be configured to be stricter in order to ensure that the retrieval result is as similar as possible to the image to be detected. If the application scenario is a commodity recommendation, the predetermined similarity threshold may be set lower, facilitating provision of a plurality of similar commodity images for user selection.
FIG. 6 shows a flow diagram of a method 600 for obtaining multiple candidate object images, in accordance with an embodiment of the present disclosure. It should be understood that method 600 may be performed, for example, at electronic device 700 depicted in fig. 7. May also be executed at the computing device 130 depicted in fig. 1. It should be understood that method 600 may also include additional acts not shown and/or may omit acts shown, as the scope of the disclosure is not limited in this respect.
At block 602, the computing device 130 determines whether the first hash value, the object class and the object attribute identification of the image to be retrieved match the identification of the hash bucket. The mark of the hash bucket is formed by splicing the image grouping information, the object category and the object attribute mark of the object image associated with the hash bucket.
If the computing device 130 determines that the first hash value, the object class and the object attribute identification of the image to be retrieved match the identification of the hash bucket, at block 604, the computing device 130 obtains a plurality of object image identifications associated with the identification of the hash bucket.
At block 606, the computing device 130 obtains image feature information for a plurality of object images associated with the plurality of object image identifications based on the plurality of object image identifications to treat the plurality of object images as a plurality of candidate object images.
In the above scheme, the first hash value, the object type of the image to be retrieved and the object attribute identifier are all matched with the identifier of the hash bucket, so that the image feature vectors of a plurality of object images associated with a plurality of object image identifiers (which are associated with the same binary image grouping information) stored in the same matching hash bucket are obtained.
FIG. 7 schematically illustrates a block diagram of an electronic device (or computing device) 700 suitable for use to implement embodiments of the present disclosure. The device 700 may be a device for implementing the method 200, 400 to 600 shown in fig. 2, 4 to 6. As shown in fig. 7, device 700 includes a Central Processing Unit (CPU) 701 that may perform various appropriate actions and processes in accordance with computer program instructions stored in a Read Only Memory (ROM) 702 or computer program instructions loaded from a storage unit 708 into a Random Access Memory (RAM) 703. In the RAM703, various programs and data required for the operation of the device 700 can also be stored. The CPU701, the ROM 702, and the RAM703 are connected to each other via a bus 704. An input/output (I/O) interface 705 is also connected to bus 704.
Various components in the device 700 are connected to the I/O interface 705, including: the input unit 706, the output unit 707, the storage unit 708, the processing unit 701 performs the various methods and processes described above, such as performing the methods 200, 400 to 600-for example, in some embodiments the methods 200, 400 to 600 may be implemented as a computer software program stored on a machine readable medium, such as the storage unit 708. In some embodiments, part or all of a computer program may be loaded onto and/or installed onto device 700 via ROM 702 and/or communications unit 709. When the computer program is loaded into the RAM703 and executed by the CPU701, one or more operations of the methods 200, 400 to 600 described above may be performed. Alternatively, in other embodiments, the CPU701 may be configured by any other suitable means (e.g., by way of firmware) to perform one or more of the acts of the methods 200, 400-600.
It should be further appreciated that the present disclosure may be embodied as methods, apparatus, systems, and/or computer program products. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied thereon for carrying out various aspects of the present disclosure.
The computer readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media as used herein is not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through a fiber optic cable), or electrical signals transmitted through electrical wires.
The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device.
The computer program instructions for carrying out operations of the present disclosure may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry that can execute the computer-readable program instructions implements aspects of the present disclosure by utilizing the state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).
Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.
These computer-readable program instructions may be provided to a processor in a voice interaction device, a processing unit of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.
The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
Having described embodiments of the present disclosure, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.
The above are merely alternative embodiments of the present disclosure and are not intended to limit the present disclosure, which may be modified and varied by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure should be included in the protection scope of the present disclosure.