CN108650517B - Based on the determination method of group's image coding multiple reference images of object - Google Patents
Based on the determination method of group's image coding multiple reference images of object Download PDFInfo
- Publication number
- CN108650517B CN108650517B CN201810465590.3A CN201810465590A CN108650517B CN 108650517 B CN108650517 B CN 108650517B CN 201810465590 A CN201810465590 A CN 201810465590A CN 108650517 B CN108650517 B CN 108650517B
- Authority
- CN
- China
- Prior art keywords
- image
- coded sequence
- image set
- sift feature
- determined
- 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
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/85—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using pre-processing or post-processing specially adapted for video compression
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
- H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Signal Processing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention proposes a kind of determination methods of group's image coding multiple reference images based on object, it is intended to improve the code efficiency of image.Implementation step are as follows: (1) obtain the Weighted Directed Graph of image set to be encoded；(2) root node of image set minimum spanning tree to be encoded is obtained；(3) it establishes and has determined that coded sequence image set and not determining coded sequence image set；(4) object detection is carried out to every image；(5) object matches are carried out to every image and residue character point matches；(6) multiple reference images for completing an image determine；(7) multiple reference images of the circulation step (5) to step (6) until having determined every image.The match point area coverage for present invention adds object detection, being matched using object matches and residue character point, and being considered in matching, takes full advantage of correlation between image, hence it is evident that improve code efficiency.It can be used for album compression, computer vision and the storage of cloud image etc..
Description
Technical field
The invention belongs to technical field of image processing, are related to a kind of groups image coded reference pictures and determine method, specifically
It is a kind of determination method of group's image coding multiple reference images based on object, can be used for album and cloud image set
Compression etc..
Background technique
In recent years, due to the appearance that cloud computing and cloud store, many large-scale scientific & technical corporation all externally provide cloud storage
Business, and many users can select image data being saved in cloud, so a large amount of image data will be will appear beyond the clouds
Storage.Cloud image data Efficient Compression storage has become an important research direction.Cloud image is using traditional at present
Single image encodes such as JPEG, MPEG, not in view of the similitude between image, so that compiling for entire cloud image set
Code efficiency is not high.In order to improve the code efficiency of image set, carrying cost is reduced, group's image coding is suggested.Existing group
Image coding mainly generates pseudo- video structure using the correlation between group's image, then is compressed using video compression technology,
The redundancy between image is made full use of, the memory space of large nuber of images in internet is dramatically saved.It is directed to group at present
The research of image coded reference relationship has single referring-to relation and two kinds of multiple reference images.Group's image encodes single referring-to relation really
It is fixed that single referring-to relation is mainly generated using minimal spanning tree algorithm, it is significantly simpler to implement, but for the redundancy of image set
Use of information is inadequate.
The determination that group's image encodes multiple reference images is mainly on the basis of single referring-to relation determines method, using repeatedly
It is constantly that every image finds new reference picture, and guarantees that the reference picture similar area found every time will not for algorithm
It repeats, can make full use of the redundancy in image set, improve group's image coding efficiency, for example, Publication No.
CN107194961A, the patent application of entitled " the determination methods of multiple reference images in group's image coding ", discloses one kind
The determination method of multiple reference images, this method generate the minimum of image set first with SIFT feature information in group's image coding
Then spanning tree determines multiple reference images on the basis of this minimum spanning tree, realize that step has for L layers: building known coded
Precedence diagram image set HE；It with Feature Points Matching method is L layers of all picture search multiple reference images in HE and L tomographic image；It will refer to
The image that image belongs to HE is added in HE；Judging L layers, whether there are also do not determine coded sequence image；Have and then reference picture is searched for
Replacement；It is added in HE with the smallest image of average weight in coded sequence image not determining after replacement；If L layers of reference picture is all
It has been determined that output L tomographic image relevant information；The above steps are repeated completes all layers of reference picture of group's image and coded sequence
It determines.This method discovers and uses the redundancy between multiple images, improves group's image by SIFT feature information iteration
Code efficiency.But it has a defect that and only considers that statement is inaccurate using SIFT matching distance for the distance between image,
And the influence that objects in images determines referring-to relation is not considered, cause last code efficiency still lower.
Summary of the invention
It is an object of the invention in view of the above-mentioned drawbacks of the prior art, proposing a kind of group's image based on object
The determination method for encoding multiple reference images, is asked for solving the lower technology of group's image coding efficiency existing in the prior art
Topic.
To achieve the above object, the technical solution that the present invention takes includes the following steps:
(1) Weighted Directed Graph of image set to be encoded is obtained:
(1a) extracts the SIFT feature of every image in image set to be encoded, and carries out SIFT feature to image two-by-two
Match, obtains multiple SIFT features matching pair of each pair of image；
(1b) calculates the SIFT average distance of multiple SIFT features matching pair of each pair of image, and will be in image set to be encoded
Every image abstraction is node, and SIFT average distance constitutes the cum rights of image set to be encoded as the weight between each pair of image
Digraph；
(2) root node of image set minimum spanning tree to be encoded is obtained:
Image set minimum spanning tree to be encoded is calculated by the Weighted Directed Graph of image set to be encoded, and to minimum spanning tree
Root node be marked；
(3) it establishes and has determined that coded sequence image set and do not determine coded sequence image set:
Image set to be encoded is divided into two sub- image sets, and the subgraph image set that will contain only minimum spanning tree root node
As having determined that coded sequence image set, using the subgraph image set being made of the residual image except removing root node as not determining
Coded sequence image set；
(4) every image for treating coded image concentration carries out object detection, obtains the multiple objects image of every image；
(5) every figure in coded sequence image set is had determined that every image in not determining coded sequence image set and currently
As carrying out object matches and the matching of residue character point:
(5a) is to every image in not determining coded sequence image set and has determined that every image in coded sequence image set
Carry out object matches, do not determined the corresponding reference picture of the multiple objects of every image in coded sequence image set with
And multiple objects and the matched SIFT feature of corresponding reference picture；
(5b) obtains the remaining SIFT feature for not determining every image in coded sequence image set:
Never determine that rejecting step (5a) obtains multiple in the SIFT feature of every image of coded sequence image set
Object and the matched SIFT feature of corresponding reference picture, are not determined the residue of every image in coded sequence image set
SIFT feature；
(5c) to do not determine coded sequence image set in every image remaining SIFT feature and have determined that coded sequence
Every image carries out repeatedly remaining SIFT feature matching in image set, and it is more accordingly to obtain every image residue SIFT feature
A reference picture；
(5d) obtains the coding weight W and corresponding multiple reference images for not determining every image in coded sequence image set:
The coding weight W for not determining every image in coded sequence image set is calculated, and by the multiple objects of every image
Multiple reference pictures of corresponding reference picture and remaining SIFT feature respectively, the multiple reference images as every image；
(6) multiple reference images for not determining minimum code weight correspondence image in coded sequence image set are obtained:
It will not determine in coded sequence image set that the corresponding image of minimum code weight is added to and have determined that coded sequence figure
In image set, and obtain the multiple reference images of minimum code weight correspondence image；
(7) judgement does not determine whether coded sequence image set is empty:
Judgement does not determine whether coded sequence image set is empty, if so, exporting in image set to be encoded the more of every image
Otherwise reference picture executes step (5)；
The present invention compared with prior art, has the advantage that
First, the present invention when obtaining group's image coding multiple reference images by carrying out object detection to every image,
And the object obtained based on detection carries out object matches to every image and residue character point matches, and reduces SIFT feature
Matching error can accurately obtain the multiple reference images of every image, take full advantage of the redundancy between image, and existing
Some technologies are compared, and the efficiency of group's image coding is effectively increased.
Second, the present invention is distance weighted by using SIFT match point area coverage and SIFT when carrying out Feature Points Matching
With indicate matched similarity, to avoid indicate that matching similarity leads to accuracy only with SIFT distance in the prior art
Low defect more accurately obtains the multiple reference images of every image, makes full use of the redundancy between image, further
Improve group's image coding efficiency.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention；
Fig. 2 is the implementation flow chart of object matches in the present invention；
Fig. 3 is the matched implementation flow chart of residue character point in the present invention.
Specific embodiment
In the following with reference to the drawings and specific embodiments, the present invention is described in further detail.
Referring to Fig.1, the determination method of a kind of groups picture coding order, includes the following steps:
Step 1) starts the Weighted Directed Graph for obtaining image set to be encoded first:
Step 1a) it treats coded image and concentrates every image zooming-out SIFT feature, using being provided in the library opencv2.4.9
Existing SIFT extract function and carry out SIFT feature extraction, and save the correspondence SIFT feature that every image zooming-out arrives；
Step 1b) to progress SIFT matching between image two-by-two, and according to the SIFT between matching result calculating two-by-two image
Matching distance, SIFT is matched and is calculated the SIFT matching distance between two images steps are as follows between two images:
Step 1b1) for two images: i image carries out SIFT matching to j image, first setting i image and j image
SIFT matches point set Fi,j；
Step 1b2) it is concentrated from the SIFT feature of image i and successively takes out not matched SIFT feature ki, with figure
As each not matched SIFT feature of j calculates SIFT feature distance value d (ki,kj), it is calculated using following formula:
Wherein vi(ki) indicate i image kthiA SIFT feature vector, vj(kj) indicate j image kthjA SIFT feature to
Amount；
Step 1b3) it is chosen and image i kth in image jiThe characteristic point k ' of a SIFT feature matching minimum rangejAway from
It is d ' from valuei(ki,k′j) and time small distance characteristic point k "jDistance value is d "i(ki,k″j), work as satisfactionWhen
It is determined as kth in i imageiKth in a SIFT feature and j image 'jA SIFT feature successful match, matching double points are saved
To set Fi,jIn, and this pair of of match point is labeled as match point, otherwise it is assumed that it fails to match, only by kth in i imageiIt is a
SIFT feature is labeled as match point；
Step 1b4) judge whether that all SIFT features have been labeled as match point in i image, it is to carry out step
1b5), no to repeat step 1b2) arrive 1b4)；
Step 1b5) calculate SIFT distance e between two images i and ji,j:
|Fi,j| indicate Fi,jThe number of matching double points, f in seti(k) and fj(k) it indicates in Fi,jK-th of matching in set
Point pair, for | F in testi,j| middle matching is set as 65535 to distance image of the number less than 20, indicates that distance is infinite
Greatly, two images are uncorrelated, i.e., finally obtain two image i images and match point set F to the SIFT of j imagei,jAnd SIFT away from
From ei,j, distance here be all it is directive, the distance of above-mentioned calculating is distance of the i image to j image, all to image set
Image two-by-two between all carry out calculating as above, the SIFT obtained between image two-by-two matches point set Fi,jWith SIFT distance ei,j, should
Every image abstraction in image set to be encoded is node, constructs figure to be encoded by distance as the weight between two images
The Weighted Directed Graph of image set；
Step 2) determines the root node of whole image collection coding:
The minimum spanning tree of image set to be encoded is generated using minimum spanning tree Zhu Liu algorithm to Weighted Directed Graph, record is most
The root node of small spanning tree；
Step 3) foundation has determined that coded sequence image set and does not determine coded sequence image set:
Image set to be encoded is divided into two subsets, using the subgraph image set for containing root node as having determined that coded sequence
Image set, another subgraph image set is as not determining coded sequence image set；
Every image that step 4) treats coded image concentration carries out object detection, obtains multiple objects image set:
About there are many object detection algorithms, YOLO object detection algorithms are selected to carry out object identification and cutting here, because
The technology that deep learning is used for YOLO algorithm has used convolutional neural networks to carry out object identification, and the fast accuracy of speed is high, will
The multiple objects that every image is cut into are put under the file of correspondence image title；
Step 5) calculates the coding of current uncoded every image apart from weight W:
Step 5a) never coded image concentrate take out an image be used as current target image, carry out object matches, acquisition
The reference picture of object matches, as shown in Fig. 2 object matches step:
Step 5a1) judge in current target image object whether all overmatching, if it is not, carrying out step 5a2), if
It is to then follow the steps 5a4)；
Step 5a2) taken out from the object of current target image one be not carried out matched subject image and it is all really
Yard sequential picture of delimiting the organizational structure carries out SIFT matching, calculates normalization object distance value Object_Dis, and calculation formula is as follows:
Wherein, Object_S is current object and have determined that present image matches multiple in coded sequence image set
SIFT feature area shared in current object, the here calculating of area use triangulation, calculate respective point
Area shared by triangulation net, Object_D are current objects and have determined that the multiple of present image in coded sequence image set
The normalization average distance value of matched SIFT feature, col × row are not determine present image in coded sequence image set
Size；
Step 5a3) select image corresponding with the matched smallest object distance value Object_Dis value of current object as
The best reference picture of the object, is recorded, and then executes step 5a1)；
Step 5a4) according to the calculated all objects of current target image the corresponding Object_Dis value of best match
It is ranked up from small to large, Num_object before selecting, and using corresponding optimal reference as the ginseng for the object matches chosen
Image is examined, the size of Num_object is set as needed, if the object number cut out is less than Num_object, Num_
Object=cuts out object number, and Num_object is set as 3 in this emulation；
Step 5b) using the characteristic point of target image rejecting object matches, obtain target image residue character point；
Step 5c) using the progress residue character point matching of target image residue character point, obtain the reference of residue character point
Image:
Step 5c1) using target image current residual match point and it is all have determined that in coded sequence image set image into
Row matching, calculates residue character point distance value Remainder_Dis
Wherein, Remainder_D be target image residue character point with currently have determined that it is corresponding in coded sequence image set
After image carries out SIFT matching, the normalized average distance value of SIFT match point, Remainder_S is that the residue of target image is special
Sign point has determined that SIFT match point is in the target image in coded sequence image set after correspondence image progress SIFT matching with current
Area coverage, col × row are the size of target image；
Step 5c2) choose the corresponding reference picture of the smallest residue character point distance value Remainder_Dis be added to it is surplus
In the reference picture of remaining characteristic point；
Step 5c3) from the residue character point of present target image reject corresponding of the smallest Remainder_Dis value
With characteristic point, using residue character point as new residue character point；
Step 5c3) judge whether new residue character point number is less than given threshold or has found Num_
The reference picture of Remainder residue character point, if so, continuing step 5d), if it is not, new residue character point is made
For current residual characteristic point, step 5c1 is executed), threshold value is set as 20, Num_Remainder in this emulation and is set as 3；
Step 5d) calculate current target image coding apart from weight W, calculated according to following formula:
Wherein, col × row indicates the current size for not determining image in coded sequence image set, D are as follows:
S are as follows:
The meaning of above-mentioned each symbol are as follows: Num_Object indicates the number of the reference picture of multiple objects, Num_
Remainder represents the number of the reference picture of residue character point, Object_DiIndicate do not determine coded sequence image set in when
The multiple SIFT matchings of i-th of the object and corresponding reference picture of preceding image are to normalization average distance, Remainder_DiTable
Show and does not determine that the corresponding normalization of the matched reference picture of i-th residue character point of present image in coded sequence image set is flat
Equal SIFT match point distance, obj_ref_matchesi.size () indicates not determine present image in coded sequence image set
The number of the SIFT of i-th of object and corresponding reference picture matching pair, remainder_ref_matchesi.size () indicates
The SIFT matching that i-th residue character point matches in the residue character point of present image in coded sequence image set is not determined
Pair number, Object_SiIndicate not determine in coded sequence image set i-th of object of present image and corresponding with reference to figure
As area coverage of the matched SIFT feature in present image, Remainder_SiIt indicates not determine coded sequence image set
Covering of the SIFT feature that i-th residue character point matches in the residue character point of middle present image in present image
Area；
Step 6) obtains the multiple reference images for not determining minimum code weight correspondence image in coded sequence image set:
It will not determine in coded sequence image set that the corresponding image of minimum code weight is added to and have determined that coded sequence figure
In image set, and obtain the multiple reference images of minimum code weight correspondence image；
Step 7) judgement does not determine whether coded sequence image set is empty:
Judgement does not determine whether coded sequence image set is empty, if so, exporting in image set to be encoded the more of every image
Otherwise reference picture executes step 5)；
Step 8) carries out more reference encoders according to the corresponding reference picture of every image, first progress geometry deformation and luminosity
Conversion, then generates multiple forecast images, using HEVC interframe encode principle, to current encoded image with multiple forecast images work
For with reference to progress interframe encode.
Claims (3)
1. a kind of determination method of group's image coding multiple reference images based on object, which comprises the steps of:
(1) Weighted Directed Graph of image set to be encoded is obtained:
(1a) obtains the SIFT feature of every image in image set to be encoded, and carries out SIFT feature matching to image two-by-two,
Obtain multiple SIFT features matching pair of each pair of image；
(1b) calculates the SIFT average distance of multiple SIFT features matching pair of each pair of image, and will be every in image set to be encoded
Image abstraction is node, and for SIFT average distance as the weight between each pair of image, the cum rights for constituting image set to be encoded is oriented
Figure；
(2) root node of image set minimum spanning tree to be encoded is obtained:
Image set minimum spanning tree to be encoded is calculated by the Weighted Directed Graph of image set to be encoded, and to the root of minimum spanning tree
Node is marked；
(3) it establishes and has determined that coded sequence image set and do not determine coded sequence image set:
Image set to be encoded is divided into two sub- image sets, and will contain only the subgraph image set of minimum spanning tree root node as
It has determined that coded sequence image set, is encoded using the subgraph image set being made of the residual image except removing root node as not determining
Precedence diagram image set；
(4) every image for treating coded image concentration carries out object detection, obtains the multiple objects image of every image；
(5) to do not determine in coded sequence image set every image and currently have determined that in coded sequence image set every image into
Row object matches and the matching of residue character point:
(5a) is to every image in not determining coded sequence image set and has determined that every image carries out in coded sequence image set
Object matches are not determined the corresponding reference picture of the multiple objects of every image in coded sequence image set and more
A object and the matched SIFT feature of corresponding reference picture；
(5b) obtains the remaining SIFT feature for not determining every image in coded sequence image set:
Never it determines and rejects the multiple objects that step (5a) is obtained in the SIFT feature of every image of coded sequence image set
With the matched SIFT feature of corresponding reference picture, do not determined that the remaining SIFT of every image in coded sequence image set is special
Sign point；
(5c) to do not determine coded sequence image set in every image remaining SIFT feature and have determined that coded sequence image
It concentrates every image to carry out repeatedly remaining SIFT feature matching, obtains every image residue SIFT feature multiple ginsengs accordingly
Examine image；
(5d) obtains the coding weight W and corresponding multiple reference images for not determining every image in coded sequence image set:
The coding weight W for not determining every image in coded sequence image set is calculated, and the multiple objects of every image are distinguished
Multiple reference pictures of corresponding reference picture and remaining SIFT feature, as the multiple reference images of every image, wherein
Encode weight W calculation formula are as follows:
Wherein, col × row indicates the size for not determining present image in coded sequence image set, and D indicates not determine coded sequence
The average reference distance value of present image in image set:
S indicates that the SIFT feature to match with all reference pictures is not determining in coded sequence image set shared by present image
The sum of area:
Num_Object indicates the number of the reference picture of multiple objects, and Num_Remainder represents the reference of residue character point
The number of image, Object_DiIt indicates not determine i-th of object of present image and corresponding reference in coded sequence image set
The multiple SIFT matchings of image are to normalization average distance, Remainder_DiIt indicates not determine and currently scheme in coded sequence image set
The matched reference picture of i-th residue character point of picture is corresponding to normalize average SIFT match point distance, obj_ref_
matchesi.size () indicates i-th of the object and corresponding reference picture that do not determine present image in coded sequence image set
SIFT matching pair number, remainder_ref_matchesi.size () indicate do not determine coded sequence image set in when
The number for the SIFT matching pair that i-th residue character point matches in the residue character point of preceding image, Object_SiIt indicates not
I-th of the object and the matched SIFT feature of corresponding reference picture for determining present image in coded sequence image set are current
Area coverage in image, Remainder_SiIt indicates not determine in coded sequence image set in the residue character point of present image
Area coverage of the SIFT feature that i-th residue character point matches in present image；
(6) multiple reference images for not determining minimum code weight correspondence image in coded sequence image set are obtained:
It will not determine in coded sequence image set that the corresponding image of minimum code weight is added to and have determined that coded sequence image set
In, and obtain the multiple reference images of minimum code weight correspondence image；
(7) judgement does not determine whether coded sequence image set is empty:
Judgement does not determine whether coded sequence image set is empty, if so, exporting in image set to be encoded more referring to for every image
Otherwise image executes step (5).
2. the determination method of group's image coding multiple reference images according to claim 1 based on object, feature exist
In to not determining every image in coded sequence image set and have determined that every in coded sequence image set described in step (5a)
It opens image and carries out object matches, realize step are as follows:
(5a1) is to each subject image in every image in not determining coded sequence image set and has determined that coded sequence image set
In every image carry out SIFT matching, do not determined each object in every image of coded sequence image set and have determined that coding
Sequential picture concentrates the SIFT feature of every images match；
(5a2) is calculated by matched SIFT feature obtained in (5a1) and is not determined every image in coded sequence image set
In each object with have determined that the matching distance value Object_Dis of every image in coded sequence image set:
Wherein, Object_S is the current object of every image and to have determined that coded sequence figure in not determining coded sequence image set
Multiple SIFT features that present image matches in image set area shared in current object, Object_D is not determining volume
Code sequential picture concentrates the current object of every image and has determined that the multiple matched of present image in coded sequence image set
The normalization average distance value of SIFT feature, col × row are the size for not determining present image in coded sequence image set；
(5a3) chooses best match distance of each object matches distance value Object_Dis minimum value as the object, corresponding
Have determined that best reference picture of the image as the object in coded sequence image set, and obtain the object and the optimal reference
The SIFT feature of images match；
(5a4) to the best match distance value for not determining all objects in every image in coded sequence image set carry out from it is small to
Big sequence selects the corresponding best match image of best match distance value of Num_object object before every image as often
It opens image multiple objects and distinguishes corresponding reference picture, corresponding matched SIFT feature is as the multiple objects of every image
Body respectively with the matched SIFT feature of corresponding reference picture.
3. the determination method of group's image coding multiple reference images according to claim 1 based on object, feature exist
In, described in step (5c) to do not determine coded sequence image set in every image remaining SIFT feature and have determined that volume
Code sequential picture concentrates every image to carry out repeatedly remaining SIFT feature matching, realizes step are as follows:
(5c1) be not by determining the remaining SIFT feature of every image in coded sequence image set and having determined that coded sequence figure
Every image carries out SIFT matching in image set, is not determined the remaining SIFT feature of every image in coded sequence image set
With the SIFT feature for having determined that every images match in coded sequence image set；
(5c2) is calculated by matched SIFT feature obtained in (5c1) and is not determined every image in coded sequence image set
Remaining SIFT feature with have determined that the matching distance Remainder_Dis of every image in coded sequence image set:
Wherein, Remainder_D is the normalization average distance value of matched SIFT feature, and Remainder_S is obtained
The corresponding area coverage for not determining the SIFT feature of present image in coded sequence image set in matched SIFT feature,
Col × row is the size for not determining present image in coded sequence image set；
(5c3), which is obtained, not to be determined the remaining SIFT feature of every image in coded sequence image set and has determined that coded sequence figure
The Remainder_Dis of all images in image set selects the wherein corresponding figure having determined that in coded sequence image set of minimum value
Picture, does not determine one of the reference picture of the remaining SIFT feature of every image in coded sequence image set as this, and records
Corresponding Remainder_D and Remainder_S；
(5c4) rejects what the reference picture that it is obtained with step (5c3) matched from the remaining SIFT feature of every image
SIFT feature, and judge whether remaining SIFT feature number is less than threshold value or has found Num_Remainder and remain
The reference picture of remaining SIFT feature does not determine every image residue SIFT feature in coded sequence image set if so, obtaining
All reference pictures otherwise using current residual SIFT feature as new remaining SIFT feature, and execute step
(5c1)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810465590.3A CN108650517B (en) | 2018-05-16 | 2018-05-16 | Based on the determination method of group's image coding multiple reference images of object |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810465590.3A CN108650517B (en) | 2018-05-16 | 2018-05-16 | Based on the determination method of group's image coding multiple reference images of object |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108650517A CN108650517A (en) | 2018-10-12 |
| CN108650517B true CN108650517B (en) | 2019-11-15 |
Family
ID=63755852
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810465590.3A Active CN108650517B (en) | 2018-05-16 | 2018-05-16 | Based on the determination method of group's image coding multiple reference images of object |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108650517B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102771125A (en) * | 2009-12-10 | 2012-11-07 | Sk电信有限公司 | Encoding / decoding method and apparatus using a tree structure |
| CN104199922A (en) * | 2014-09-01 | 2014-12-10 | 中国科学院自动化研究所 | Large-scale image library retrieval method based on local similarity hash algorithm |
| CN107194961A (en) * | 2017-05-19 | 2017-09-22 | 西安电子科技大学 | The determination method of multiple reference images in colony's Image Coding |
| CN107426580A (en) * | 2017-05-16 | 2017-12-01 | 西安电子科技大学 | Colony's picture coding structure generation method based on Zhu Liu method and ant colony method |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20130037162A (en) * | 2011-10-05 | 2013-04-15 | 한국전자통신연구원 | Method and apparatus of inter-layer texture prediction for scalable video coding |
| US9111329B2 (en) * | 2011-12-26 | 2015-08-18 | Samsung Electronics Co., Ltd | Method and apparatus for compressing images |
| CN107820095B (en) * | 2016-09-14 | 2020-01-03 | 北京金山云网络技术有限公司 | Long-term reference image selection method and device |
-
2018
- 2018-05-16 CN CN201810465590.3A patent/CN108650517B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102771125A (en) * | 2009-12-10 | 2012-11-07 | Sk电信有限公司 | Encoding / decoding method and apparatus using a tree structure |
| CN104199922A (en) * | 2014-09-01 | 2014-12-10 | 中国科学院自动化研究所 | Large-scale image library retrieval method based on local similarity hash algorithm |
| CN107426580A (en) * | 2017-05-16 | 2017-12-01 | 西安电子科技大学 | Colony's picture coding structure generation method based on Zhu Liu method and ant colony method |
| CN107194961A (en) * | 2017-05-19 | 2017-09-22 | 西安电子科技大学 | The determination method of multiple reference images in colony's Image Coding |
Non-Patent Citations (1)
| Title |
|---|
| Personal photo album compression and management;Ruobing Zou et al;《2013 IEEE International Symposium on Circuits and Systems (ISCAS2013)》;IEEE;20130523;第1428-1431页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108650517A (en) | 2018-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110263659B (en) | Finger vein recognition method and system based on triplet loss and lightweight network | |
| Lin et al. | Bsn: Boundary sensitive network for temporal action proposal generation | |
| CN111177446B (en) | Method for searching footprint image | |
| CN112132197B (en) | Model training, image processing method, device, computer equipment and storage medium | |
| CN108280187B (en) | Hierarchical image retrieval method based on depth features of convolutional neural network | |
| CN111241345A (en) | Video retrieval method and device, electronic equipment and storage medium | |
| Wang et al. | Multiscale deep alternative neural network for large-scale video classification | |
| CN111931953A (en) | Multi-scale characteristic depth forest identification method for waste mobile phones | |
| Hou et al. | Airborne infrared aircraft target detection algorithm based on YOLOv4-tiny | |
| Wang et al. | Generative adversarial network based on resnet for conditional image restoration | |
| Qu et al. | Low illumination enhancement for object detection in self-driving | |
| CN114067128A (en) | SLAM loop detection method based on semantic features | |
| CN110110120B (en) | Image retrieval method and device based on deep learning | |
| CN114495170A (en) | Pedestrian re-identification method and system based on local self-attention inhibition | |
| CN111291785A (en) | Target detection method, device, equipment and storage medium | |
| CN104778272B (en) | A kind of picture position method of estimation excavated based on region with space encoding | |
| Li et al. | Detection and identification of peach leaf diseases based on YOLO v5 improved model | |
| Wang et al. | A transformer-based mask R-CNN for tomato detection and segmentation | |
| CN114329031A (en) | Fine-grained bird image retrieval method based on graph neural network and deep hash | |
| Zhao et al. | Underwater fish detection in sonar image based on an improved Faster RCNN | |
| CN108764258A (en) | A kind of optimum image collection choosing method being inserted into for group's image | |
| CN108650517B (en) | Based on the determination method of group's image coding multiple reference images of object | |
| CN110516640A (en) | It is a kind of to combine the vehicle indicated discrimination method again based on feature pyramid | |
| Cao et al. | Image retrieval via gated multiscale NetVLAD for social media applications | |
| Lin et al. | Fusion and Inpainting: A Novel Salient Object Detection Network |
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 |