CN109684493B - Image reordering method based on k neighborhood distribution score - Google Patents
Image reordering method based on k neighborhood distribution score Download PDFInfo
- Publication number
- CN109684493B CN109684493B CN201910009038.8A CN201910009038A CN109684493B CN 109684493 B CN109684493 B CN 109684493B CN 201910009038 A CN201910009038 A CN 201910009038A CN 109684493 B CN109684493 B CN 109684493B
- Authority
- CN
- China
- Prior art keywords
- image
- score
- list
- initial
- new
- 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.)
- Active
Links
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an image reordering method based on k neighborhood distribution scores, which is based on an initial ordering list for descending ordering according to image similarity and then carries out reordering according to the height of the k neighborhood distribution scores; including obtaining an initial sorted list; establishing an extended query set of a test image p; taking the initial sorted list as a new test image set; adding p to the initial reference image set G; calculating a corresponding temporary ordered list for each new test image; searching the sorting position L of k images in each temporary sorting list in the expanded query set; calculating the following scores for each reference image in the initial ranked list: a location score, an overlap score, and a dispersion score; constructing k field distribution scores of all reference images in the initial ranking list; and reordering according to the k-domain distribution scores of the reference images in the initial ranking list. The method can reduce the interference of the negative samples on the sequencing and improve the accuracy of image retrieval.
Description
Technical Field
The invention relates to the field of image retrieval, in particular to a method for sequencing images.
Background
Image retrieval is mainly based on image similarity judgment: by calculating the similarity between the image to be detected and each reference image in the reference image data set and then sequencing according to the similarity, the reference image with the highest similarity to the image to be detected is generally taken as top-1, and the accuracy of the top-1 plays a crucial role in the accuracy of the image retrieval result. However, because the reference image data set has both positive and negative samples, the negative samples interfere with the image similarity calculation, and because the positive samples have problems of photographing angle, shielding, etc., the similarity between the negative samples at certain angles or without shielding and the image to be measured is higher than the similarity between the positive samples and the image to be measured, so that the sequencing obtained by adopting the prior art and relying on the similarity calculation is inaccurate, and even the negative samples are arranged to top-1.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides an image reordering method based on k neighborhood distribution scores, which solves the technical problem that ordering is performed by simply depending on image similarity and is easily interfered by negative samples, can reduce the interference of the negative samples on ordering, and improves the accuracy of image retrieval.
In order to solve the technical problems, the invention adopts the following technical scheme: an image reordering method based on k neighborhood distribution scores is characterized in that reordering is carried out according to the height of the k neighborhood distribution scores on the basis of an initial ranking list which is subjected to descending ranking according to image similarity; the method comprises the following steps:
step 1: acquiring an initial reference image set G-G containing N reference images i1,2,3, and N, calculating the similarity of a single test image p and each reference image, and sorting according to the image similarity from high to low, thereby obtaining an initial sorted listWherein the content of the first and second substances,is a reference picture ranked at the ith position in the initial ranking list;
step 2: an extended query set Q (p, k) of test images p is created, as follows:
Q(p,k)={p}+N(p,k-1);
where p is the test image, N (p, k-1) is from the first k-1 reference images in the initial ordered list omega (p, G),
and step 3: will initially rank the listAs a new test image setAnd the number of the first and second electrodes,
and 4, step 4: adding a test image p to an initial referenceIn the image set G, a new reference image set G is obtainednew={p}+G;
And 5: traversing a new test image set PnewCalculating by PnewIn turn, as new test images to a new reference image set GnewThe similarity of each reference image is determined, and a new reference image set G is set according to the similarity from high to lownewThe reference images are sorted, so that a temporary sorting list is obtained corresponding to each new test image;
step 6: searching the ranking positions L of k images in the expanded query set Q (p, k) in each temporary ranking list according to the temporary ranking list corresponding to each new test image;
and 7: the following scores were calculated for each reference image in the initial ranked list Ω (p, G): location score Sd-lOverlap score Sd-oAnd a discrete score Sd-d(ii) a Wherein the content of the first and second substances,the scores of the terms are respectively calculated according to the following formula:
wherein the test image p in the extended query set Q (p, k) is in the new test imageThe corresponding sorting position in the temporary sorting list is L (p); expanding reference images in query set Q (p, k)In the new test imageThe corresponding sorting position in the temporary sorting list isThe weight coefficient of the test image p isk-1 reference imagesAll the weight coefficients of
Wherein N (p, k) represents the nearest k fields of the test image p, i.e., the image set composed of the first k reference images in the initial sorted list Ω (p, G);representing new test imagesThe latest k field of (i.e. new test image)An image set consisting of the first k reference images in the corresponding temporary ordered list; card []Represents the calculation of N (p, k) andthe number of identical images;
wherein var [. cndot. ] represents variance calculation;
and 8: constructing k-region distribution scores S of each reference image in each initial ranking list omega (p, G)dAccording to any one of the following modes:
taking the position score as a k-domain distribution score: sd=Sd-l;
Taking the overlap score as a k neighborhood distribution score: sd=Sd-o
Taking the product of the position score and the overlap score as the k neighborhood distribution score: sd=Sd-l×Sd-o;
Taking the product of the discrete score and the overlap score as a k neighborhood distribution score: sd=Sd-d×Sd-o;
Taking the product of the position score, the discrete score and the overlap score as a k neighborhood distribution score: sd=Sd-l×Sd-d×Sd-o;
And step 9: and reordering the reference images in the initial sorted list omega (p, G) from high to low according to the k domain distribution scores of the reference images in the initial sorted list omega (p, G), so as to obtain a reordered list based on the k neighborhood distribution scores.
Preferably, the characteristic distance is used as a similarity judgment index of the test image and the reference image.
Preferably, the value k is determined according to the average number n of reference images owned by the same target object in the initial reference image set G, and the value range of k is [0.6n,0.7n ].
Compared with the prior art, the invention has the following beneficial effects:
1. according to the image retrieval method, on the basis of the initial ranking list which is subjected to descending ranking according to the image similarity, the distribution scores of all the reference images according to the k neighborhood are calculated, and the re-ranking is performed according to the distribution scores of the k neighborhood, so that the defect that negative sample interference cannot be avoided due to the fact that the ranking is performed only by depending on the image similarity is overcome, the interference of the negative samples on the ranking can be reduced, and the accuracy of image retrieval is improved.
2. The invention provides 5 kinds of structure k field distribution scores SdThe method can improve the accuracy rate on the basis of the initial ranking list. Wherein, especially when Sd=Sd-l×Sd-d×Sd-oThe highest accuracy is obtained.
3. The invention adopts the characteristic distance as the similarity judgment index, such as Euclidean distance, cosine distance and the like in the prior art.
4. According to the invention, the k value is determined according to the composition of the initial reference image data set, and according to a large number of experimental statistics, when the value interval of k is [0.6n,0.7n ], reordering can obtain the highest accuracy.
5. The reordering method has the advantages of simple calculation and low complexity, and can realize efficient reordering.
Drawings
FIG. 1 is a reference schematic diagram of a temporary ordered list corresponding to each new test image in step 5;
FIG. 2 is a schematic diagram of the calculation of the overlap score;
FIG. 3 is a schematic diagram illustrating the effect of the image reordering method based on k neighborhood distribution scores;
FIG. 4 is a graph of a re-identification performance evaluation of the image data set CUHK03 by changing the value of k;
fig. 5 is a re-recognition performance evaluation graph of the image data set Market1501 by changing the k value.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and preferred embodiments.
An image reordering method based on k neighborhood distribution scores is characterized in that reordering is carried out according to the k neighborhood distribution scores on the basis of an initial ranking list which is subjected to descending ranking according to image similarity; the method comprises the following steps:
step 1: acquiring an initial reference image set G-G containing N reference images i1,2,3, and N, calculating the similarity of a single test image p and each reference image, and sorting according to the image similarity from high to low, thereby obtaining an initial sorted listWherein the content of the first and second substances,is the reference picture ranked at the ith position in the initial ranking list.
Step 2: an extended query set Q (p, k) of test images p is created, as follows:
Q(p,k)={p}+N(p,k-1);
where p is the test image, N (p, k-1) is from the first k-1 reference images in the initial ordered list omega (p, G),
and step 3: will initially rank the listAs a new test image setAnd the number of the first and second electrodes,
and 4, step 4: adding the test image p to the initial reference image set G to obtain a new reference image set Gnew={p}+G。
And 5: traversing a new test image set PnewCalculating by PnewIn turn, as new test images to a new reference image set GnewThe similarity of each reference image is determined, and a new reference image set G is set according to the similarity from high to lownewSo as to obtain a temporary ordered list corresponding to each new test image, as shown in fig. 1, corresponding to each new test imageNew reference image set GnewEach reference image in the image set is newly ranked, and I in FIG. 1 represents the image from the new self-reference image set GnewBut not images of the extended query set Q (p, k) of p.
Step 6: and searching the ranking positions L of the k images in the expanded query set Q (p, k) in each temporary ranking list according to the temporary ranking list corresponding to each new test image, wherein the ranking positions are sequence numbers in the ranking, and the reciprocal of the ranking positions is the score of the ranking positions.
And 7: the following scores were calculated for each reference image in the initial ranked list Ω (p, G): location score Sd-lOverlap score Sd-oAnd a discrete score Sd-d(ii) a Wherein the content of the first and second substances,the scores of the terms are respectively calculated according to the following formula:
wherein the test image p in the extended query set Q (p, k) is in the new test imageThe corresponding sorting position in the temporary sorting list is L (p); expanding reference images in query set Q (p, k)In the new test imageThe corresponding sorting position in the temporary sorting list isThe weight coefficient of the test image p isk-1 reference imagesAll the weight coefficients of
The calculation principle of the overlap score is shown in fig. 2:
n (p, k) represents the nearest k fields of the test image p, i.e., the image set composed of the first k reference images in the initial sorted list Ω (p, G);representing new test imagesThe latest k field of (i.e. new test image)An image set consisting of the first k reference images in the corresponding temporary ordered list; card []Represents the calculation of N (p, k) andthe number of identical images;
where var [. cndot. ] represents the variance calculation.
And 8: constructing k-region distribution scores S of each reference image in each initial ranking list omega (p, G)dAccording to any one of the following modes:
taking the position score as a k-domain distribution score: sd=Sd-l;
Taking the overlap score as a k neighborhood distribution score: sd=Sd-o
Taking the product of the position score and the overlap score as the k neighborhood distribution score: sd=Sd-l×Sd-o;
Taking the product of the discrete score and the overlap score as a k neighborhood distribution score: sd=Sd-d×Sd-o;
Taking the product of the position score, the discrete score and the overlap score as a k neighborhood distribution score: sd=Sd-l×Sd-d×Sd-o。
And step 9: and reordering the reference images in the initial sorted list omega (p, G) from high to low according to the k domain distribution scores of the reference images in the initial sorted list omega (p, G), so as to obtain a reordered list based on the k neighborhood distribution scores.
The effect of this is seen in fig. 3, setting k to 5, top, test image P and the first 9 samples in the initial ranking table, where N1-N2 are negative samples and P1-P7 are positive samples, the query expansion set Q (P,5) for test image P consists of P, N1, P1, P2 and N2, then setting N1 to the new test image, middle, nearest neighbor N (N1,5) and Q (P,5) of image N1 in the new ranking table 5 position tags, 39,1,504,437 and 47 respectively, resulting in a distribution and a dispersion score of 0.141 and 4.11, 48-3, Q (P,5) and N (N1,5) identical images, so the number of overlapping is 1, the three scores are multiplied together to give a final score of 0.58-10, 3, the remaining images are found by the same method of × -3, the bottom score of P is found to be at the lower score of P2, P2, P2 is found to be at the bottom of the ranking table according to the final score of P, P2, P2, and P2.
In order to further verify the effect of the invention, the characteristic distance is used as an image similarity judgment index, and the proposed method is evaluated on two large data sets: CUHK03 and Market 1501.
CUHK03 consists of 13164 images for a total of 1467 pedestrians, collected by two different cameras, including a manually labeled bounding box and a bounding box detected by a Deformable Part Model (DPM), which is used herein in a single shot mode, the data set can be divided into a training set containing 1367 people and a test set containing 100 people, images are selected from the second camera as the test set, and one image is randomly selected from the images from the first camera view for each pedestrian to form a reference image set.
Market1501 contains 32668 images of 1501 pedestrians from six cameras, divided into two parts: 12,936 images from 751 pedestrians were used as training sets and 19,732 images from 750 pedestrians were used as test sets, with DPM to detect bounding boxes. A similar test protocol was used as with the CUHK03 data set.
Image reordering method based on k-domain distribution for overall evaluation
In a specific embodiment, the proposed image reordering method based on k-domain distribution is compared with other existing reordering methods, and the recognition performance without using any reordering method is taken as a reference line, as shown in table 1:
TABLE 1
Reordering method | CUHK03 | Market1501 |
Reference line | 91.2 | 82.9 |
CDM | 91.5 | 83.3 |
AQE | 91.3 | 83.1 |
SCA | 92.0 | 83.5 |
k-NN | 91.9 | 83.4 |
k-reciprocal neighbors | 92.1 | 84.1 |
k distribution reordering | 93.5 | 85.7 |
Context Dissimilarity Measure (CDM), Average Query Extension (AQE), Sparse Context Activation (SCA), k-Nearest Neighbor reordering (k-Nearest Neighbor-ranking, k-NN), and k-reciprocal encoding (k-reciprocal encoding) are compared with the method of the present invention. The experimental results are shown in table 1, and the results show that the reordering method of the present invention can achieve effective top-1 accuracy improvement, the baseline of CUHK03 and Market1501 data sets are 91.2% and 82.9%, respectively, and the values of k corresponding to the two data sets are 7 and 17, and reordering with k distribution can achieve 2.3% and 2.8% improvement, which can be found to exceed other methods.
(II) different construction modes for evaluating k field distribution scores
It has been mentioned previously that the k-distribution reordering score may be comprised of three components, a position score, a discrete score and an overlap score, and thus these three separate scores and from their pairwise combination result in the acquisition of three additional potential scores, "position score × discrete score", "position score × overlap score" and "discrete score × overlap score" may constitute six score types, taking the data set CUHK03 as an example, the experimental results for each score are shown in table 2.
TABLE 2
Score type | Top-1 |
Location scoring | 92.9 |
Discrete score | 6.7 |
Overlap score | 93.2 |
Location score × discrete score | 84.9 |
Location score × overlap score | 92.6 |
Discrete score × overlap score | 92.9 |
Location score × discrete score × overlap score | 93.5 |
It was found that the score comprising the three parts achieved the best results, better than the other score combinations, 0.3% and 0.6% higher than the single type "overlap score" and the combined type "discrete score × overlap score" for best performance respectively-it is noted that the effect based on "discrete score" was only poor, reaching 6.7% and decreased more than the baseline 91.2%, but with this score discarded, the "position score × overlap score" combination achieved a top-1 accuracy of 92.6% and a 0.9% reduction over the combination containing the three scores, thus combining it with the "position score" and "overlap score" to achieve complementation of different attribute information and ultimately the best performance improvement.
(III) evaluation of the influence of the k-value
Influence of parameters: in the image reordering method of k-domain distribution score, k value is variable, and in all the graphs presented before, corresponding to two data sets with k-7 and k-17, considering that the composition of each data set is different, the influence of changing k value on the re-recognition performance is evaluated. As shown in fig. 4 and 5, it is found that when the value of the parameter k is within the intervals [6,14] and [9,23] on the two data sets, the performance is better than the baseline, the improvement is 0.8 to 2.3% for CUHK03, the best top-1 accuracy is achieved to 93.5% when k is 7, the improvement is 0.5 to 2.8% for mark 1501, and the highest accuracy of 85.7% can be obtained when k is 17. The average number of the same pedestrian in the reference image sets of the two data sets is 9.76 and 26.3, and it can be seen that when the value of k is 0.6-0.7 times of the value, the proposed reordering strategy can realize higher re-identification performance. Moreover, when k exceeds a certain threshold, top-1 accuracy will gradually decrease, mainly because a larger parameter k will introduce more negative samples and increase errors, so as to decrease performance and increase computational complexity.
In summary, the image reordering method based on k neighborhood distribution scores of the present invention can improve the score of the positive sample by constructing the k neighborhood distribution scores and selecting the appropriate value of k, thereby reducing the interference of negative samples on reordering, and it is unsupervised and can adapt to various image retrieval tasks.
Claims (3)
1. An image reordering method based on k neighborhood distribution score is characterized in that: on the basis of an initial ranking list which is subjected to descending ranking according to the image similarity, reordering according to the k neighborhood distribution score; the method comprises the following steps:
step 1: acquiring an initial reference image set G-G containing N reference imagesi1,2,3, and N, calculating the similarity of a single test image p and each reference image, and sorting according to the image similarity from high to low, thereby obtaining an initial sorted listWherein the content of the first and second substances,is a reference picture ranked at the ith position in the initial ranking list;
step 2: an extended query set Q (p, k) of test images p is created, as follows:
Q(p,k)={p}+N(p,k-1);
where p is the test image, N (p, k-1) is from the first k-1 reference images in the initial ordered list omega (p, G),
and step 3: will initially rank the listAs a new test image setAnd the number of the first and second electrodes,
and 4, step 4: adding the test image p to the initial reference image set G to obtain a new reference image set Gnew={p}+G;
And 5: traversing a new test image set PnewCalculating by PnewIn turn, as new test images to a new reference image set GnewThe similarity of each reference image is determined, and a new reference image set G is set according to the similarity from high to lownewThe reference images are sorted, so that a temporary sorting list is obtained corresponding to each new test image;
step 6: searching the ranking positions L of k images in the expanded query set Q (p, k) in each temporary ranking list according to the temporary ranking list corresponding to each new test image;
and 7: the following scores were calculated for each reference image in the initial ranked list Ω (p, G): location score Sd-lOverlap score Sd-oAnd a discrete score Sd-d(ii) a Wherein the content of the first and second substances,the scores of the terms are respectively calculated according to the following formula:
wherein the test image p in the extended query set Q (p, k) is in the new test imageThe corresponding sorting position in the temporary sorting list is L (p); expanding reference images in query set Q (p, k)In the new test imageThe corresponding sorting position in the temporary sorting list isThe weight coefficient of the test image p isk-1 reference imagesAll the weight coefficients of
Where N (p, k) represents the nearest k fields of the test image p, i.e., the image set composed of the first k reference images in the initial sorted list Ω (p, G);Representing new test imagesThe latest k field of (i.e. new test image)An image set consisting of the first k reference images in the corresponding temporary ordered list; card []Represents the calculation of N (p, k) andthe number of identical images;
wherein var [. cndot. ] represents variance calculation;
and 8: constructing k-region distribution scores S of each reference image in each initial ranking list omega (p, G)dAccording to any one of the following modes:
taking the position score as a k-domain distribution score: sd=Sd-l;
Taking the overlap score as a k neighborhood distribution score: sd=Sd-o
Taking the product of the position score and the overlap score as the k neighborhood distribution score: sd=Sd-l×Sd-o;
Taking the product of the discrete score and the overlap score as a k neighborhood distribution score: sd=Sd-d×Sd-o;
Taking the product of the position score, the discrete score and the overlap score as a k neighborhood distribution score: sd=Sd-l×Sd-d×Sd-o;
And step 9: and reordering the reference images in the initial sorted list omega (p, G) from high to low according to the k domain distribution scores of the reference images in the initial sorted list omega (p, G), so as to obtain a reordered list based on the k neighborhood distribution scores.
2. The method of claim 1, wherein the image reordering based on k-neighborhood distribution score comprises: and adopting the characteristic distance as a similarity judgment index of the test image and the reference image.
3. The method of claim 1, wherein the image reordering based on k-neighborhood distribution score comprises: and k value is determined according to the average reference image number n of the same target object in the initial reference image set G, and the value range of k is [0.6n,0.7n ].
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910009038.8A CN109684493B (en) | 2019-01-04 | 2019-01-04 | Image reordering method based on k neighborhood distribution score |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910009038.8A CN109684493B (en) | 2019-01-04 | 2019-01-04 | Image reordering method based on k neighborhood distribution score |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109684493A CN109684493A (en) | 2019-04-26 |
CN109684493B true CN109684493B (en) | 2020-06-23 |
Family
ID=66192524
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910009038.8A Active CN109684493B (en) | 2019-01-04 | 2019-01-04 | Image reordering method based on k neighborhood distribution score |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109684493B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110502964B (en) * | 2019-05-21 | 2021-09-28 | 杭州电子科技大学 | Unsupervised data-driven pedestrian re-identification method |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103325122B (en) * | 2013-07-03 | 2016-01-20 | 武汉大学 | Based on the pedestrian retrieval method of Bidirectional sort |
WO2015022020A1 (en) * | 2013-08-13 | 2015-02-19 | Logograb Limited | Recognition process of an object in a query image |
CN107506401A (en) * | 2017-08-02 | 2017-12-22 | 大连理工大学 | A kind of image retrieval rearrangement method based on drawing method |
CN107506703B (en) * | 2017-08-09 | 2020-08-25 | 中国科学院大学 | Pedestrian re-identification method based on unsupervised local metric learning and reordering |
-
2019
- 2019-01-04 CN CN201910009038.8A patent/CN109684493B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN109684493A (en) | 2019-04-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106096561B (en) | Infrared pedestrian detection method based on image block deep learning features | |
US8892542B2 (en) | Contextual weighting and efficient re-ranking for vocabulary tree based image retrieval | |
CN108921083B (en) | Illegal mobile vendor identification method based on deep learning target detection | |
US20120301014A1 (en) | Learning to rank local interest points | |
CN104820718B (en) | Image classification and search method based on geographic location feature Yu overall Vision feature | |
CN109740541A (en) | A kind of pedestrian weight identifying system and method | |
CN109961051A (en) | A kind of pedestrian's recognition methods again extracted based on cluster and blocking characteristic | |
CN103325122B (en) | Based on the pedestrian retrieval method of Bidirectional sort | |
CN107315795B (en) | The instance of video search method and system of joint particular persons and scene | |
CN107103326A (en) | The collaboration conspicuousness detection method clustered based on super-pixel | |
CN108764269A (en) | A kind of cross datasets pedestrian recognition methods again based on space-time restriction incremental learning | |
CN101794384A (en) | Shooting action identification method based on human body skeleton map extraction and grouping motion diagram inquiry | |
Meng et al. | Object instance search in videos via spatio-temporal trajectory discovery | |
CN111160396B (en) | Hyperspectral image classification method of graph convolution network based on multi-graph structure | |
CN102890700A (en) | Method for retrieving similar video clips based on sports competition videos | |
CN110866134B (en) | Image retrieval-oriented distribution consistency keeping metric learning method | |
JP5685324B2 (en) | Method and apparatus for comparing pictures | |
CN104298992A (en) | Self-adaptive scale pedestrian re-identification method based on data driving | |
CN106776950B (en) | On-site shoe-print trace pattern image retrieval method based on expert experience guidance | |
CN104994366A (en) | FCM video key frame extracting method based on feature weighing | |
CN106295532A (en) | A kind of human motion recognition method in video image | |
CN109299664A (en) | A kind of method for reordering that pedestrian identifies again | |
CN104484679B (en) | Non- standard rifle shooting warhead mark image automatic identifying method | |
CN109684493B (en) | Image reordering method based on k neighborhood distribution score | |
CN107506429A (en) | A kind of image rearrangement sequence method integrated based on marking area and similitude |
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 |