CN102375990B - Method and equipment for processing images - Google Patents
Method and equipment for processing images Download PDFInfo
- Publication number
- CN102375990B CN102375990B CN201010257628.1A CN201010257628A CN102375990B CN 102375990 B CN102375990 B CN 102375990B CN 201010257628 A CN201010257628 A CN 201010257628A CN 102375990 B CN102375990 B CN 102375990B
- Authority
- CN
- China
- Prior art keywords
- image block
- image
- neighbour
- arest neighbors
- search
- 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.)
- Expired - Fee Related
Links
Abstract
The invention provides a method and equipment for processing images. The method for processing the images comprises the following steps of: searching one or more neighboring image blocks of all the image blocks from an image block set of a plurality of image blocks which are extracted from a sample image, wherein the one or more neighboring image blocks comprise one or more image blocks, which are the closest to all the image blocks, in the image block set; forming a neighboring set of all the image blocks; and then, storing the image block set and information of the neighboring set related to all the image blocks.
Description
Technical field
The disclosure relates to image processing field, in particular to the method and apparatus of image procossing.
Background technology
In the image processing method based on study, usually need the training set (or dictionary) being obtained image or image block by training great amount of images sample, use in order to follow-up image procossing.Such as, by training a large amount of low resolution corresponded to each other and high-definition picture sample, can obtain the training set comprising the multipair low-resolution image block that corresponds to each other and high-definition picture block, such training set can as co-occurrence patterns (Co-occurrence Model) or dictionary for utilizing the image of low resolution in the image super-resolution application generating high-definition picture.
Summary of the invention
Provide hereinafter about brief overview of the present disclosure, to provide about the basic comprehension in some of the present disclosure.Should be appreciated that this general introduction is not summarize about exhaustive of the present disclosure.It is not that intention determines key of the present disclosure or pith, neither intended limitation the scope of the present disclosure.Its object is only provide some concept in simplified form, in this, as the preorder in greater detail discussed after a while.
According to embodiments more of the present disclosure, one or more neighbour's image block of each image block can be searched in the image block set comprising the multiple image blocks extracted from sample image.These one or more neighbour's image blocks comprise one or more nearest image block of this each image block of described image block set middle distance, thus form neighbour's collection of this each image block.Then, the information described image block set can preserved and collect about the neighbour of each image block.Wherein, the multiple neighbour's image blocks searching for each image block comprise: stochastic generation proper vector, and wherein the distance of this proper vector and this image block is in predetermined scope; The arest neighbors image block that distance feature vector is nearest is searched in image block set, if the arest neighbors image block found is different from this image block, then increase the count value associated with the arest neighbors image block that this finds, wherein, stochastic generation at least two proper vectors search the arest neighbors image block of each proper vector; And by find, one or more arest neighbors image block of associating with larger count value, the neighbour as this image block collects.
According to other embodiments of the present disclosure, coarse search can be carried out in the image block set comprising the multiple image blocks extracted from sample image, obtain the image block approximate nearest with query object distance.Then, can coarse search to the neighbour of image block concentrate and search one or more image block nearest with described query object, as one or more arest neighbors image block of described query object.Wherein, coarse search to image block neighbour collection can comprise one or more nearest image block of described this image block of image block set middle distance.
According to an embodiment of the present disclosure, a kind of image processing equipment comprises image block set treating apparatus and information-storing device.Image block set treating apparatus is used in the image block set comprising the multiple image blocks extracted from sample image, searches for one or more neighbour's image block of each image block.Wherein, one or more neighbour's image block comprises one or more nearest image block of this each image block of image block set middle distance, thus forms neighbour's collection of this each image block.Information-storing device is for the information of preserving image block set and collect about the neighbour of each image block.Wherein, image block set treating apparatus is arranged to by the following multiple neighbour's image blocks searching for each image block: stochastic generation proper vector, and wherein the distance of this proper vector and this image block is in predetermined scope; The arest neighbors image block that distance feature vector is nearest is searched in image block set, if the arest neighbors image block found is different from this image block, then increase the count value associated with the arest neighbors image block that this finds, wherein, stochastic generation at least two proper vectors search the arest neighbors image block of each proper vector; And by find, one or more arest neighbors image block of associating with larger count value, the neighbour as this image block collects.
Accompanying drawing explanation
With reference to below in conjunction with the explanation of accompanying drawing to disclosure embodiment, above and other objects, features and advantages of the present disclosure can be understood more easily.Parts in accompanying drawing are just in order to illustrate principle of the present disclosure.In the accompanying drawings, same or similar technical characteristic or parts will adopt same or similar Reference numeral to represent.
Fig. 1 shows the indicative flowchart of the image processing method according to an embodiment of the present disclosure;
Fig. 2 shows the indicative flowchart of the image processing method according to another embodiment of the present disclosure;
Fig. 3 shows according to a specific embodiment of the present disclosure, the indicative flowchart searching for the method for neighbour's collection of each image block in image block set;
Fig. 4 shows according to a concrete example of the present disclosure, schematic diagram image block set recursive subdivision being become the process of multiple subset;
Fig. 5 shows the schematic diagram being organized into the image block set of tree construction according to a concrete example of the present disclosure;
Fig. 6 shows and image block set is divided into the actual divisional plane of two subsets and the schematic diagram of desirable divisional plane;
Fig. 7 shows the schematic diagram setting up pseudo-point list according to a specific embodiment of the present disclosure;
Fig. 8 shows the indicative flowchart of the image processing method according to an embodiment of the present disclosure;
Fig. 9 shows the indicative flowchart of the image processing method according to another embodiment of the present disclosure;
Figure 10 shows the schematic block diagram of the image processing equipment according to an embodiment of the present disclosure;
Figure 11 shows the schematic block diagram of the image processing equipment according to another embodiment of the present disclosure;
Figure 12 shows the schematic block diagram of the image processing equipment according to another embodiment of the present disclosure;
Figure 13 shows 4 width images, wherein (a) represents a width low-resolution image, (d) expression actual photographed, corresponding with image (a) high-definition picture, b () expression utilizes that bilinear interpolation method generates, corresponding with image (a) high-definition picture, (c) represents that utilization generates according to the method for embodiment of the present disclosure or equipment, corresponding with image (a) high-definition picture; And
Figure 14 shows and can be used for implementing the schematic block diagram according to the computing machine of embodiment of the present disclosure.
Embodiment
With reference to the accompanying drawings embodiment of the present disclosure is described.The element described in an accompanying drawing of the present disclosure or a kind of embodiment and feature can combine with the element shown in one or more other accompanying drawing or embodiment and feature.It should be noted that for purposes of clarity, accompanying drawing and eliminate expression and the description of parts that have nothing to do with the disclosure, known to persons of ordinary skill in the art and process in illustrating.
Fig. 1 shows the indicative flowchart of the image processing method according to an embodiment of the present disclosure.Wherein, after extracted image block set (or being called training set or dictionary) by the multiple sample image of training, neighbour's image block of each image block in this set is searched in image block set, thus generate neighbour's collection of each image block, so that inquiry uses in follow-up image processing process.
As shown in Figure 1, the method can comprise step 103 and 109.
In step 103, one or more image block that each image block in the image block set comprising the multiple image blocks extracted from sample image in this set of detection range is nearest.The image block searched is called one or more neighbour's image block of each image block.These neighbour's image blocks form neighbour's collection of each image block.
In the disclosure, described image block set can be the training set obtained by adopting any method to train great amount of samples image, such as, can each image block in analyzing samples image local energy and only retain the high image block of the energy corresponding with significant image primitive (as edge, turning and/or spot etc.), in this no limit and detailed description.
In step 109, this image block set can be preserved and collect relevant information with the neighbour of each image block in this set.
Utilizing the method shown in Fig. 1, after the training set obtaining comprising multiple image block by training, neighbour's collection of each image block in this training set can also be generated.While this training set of preservation, the information of neighbour's collection of wherein each image block can also be preserved.Like this, when needing to inquire about this training set in follow-up image procossing, after initial search to certain image block in training set, can also concentrate as required and its neighbour and carry out further search refinement, thus the image block more suited the requirements.Train the method gathered to compare with preserving with only generating, the method shown in Fig. 1 is more conducive to follow-up inquiry, can inquire about and obtain information more accurately.
Fig. 2 shows the indicative flowchart of the image processing method according to another embodiment of the present disclosure.Method shown in this embodiment to Fig. 1 is similar, and difference is, in the method shown in Fig. 2, image block set can be organized into the tree structure of classification.
As shown in Figure 2, the method can comprise step 201,203 and 209.Particularly, in step 201, first the image block set comprising the multiple image blocks extracted from sample image can be organized into multistage tree construction.In this tree construction, root node can correspondence image set of blocks, and comprises the information of the thresholding for this set-partition being become the subset corresponding with next stage node.A subset of each branch node correspondence image set of blocks also comprises the information of the thresholding for the subset corresponding with this branch node being divided into the subset corresponding with next stage node; And or one group of image block in each leaf node correspondence image set of blocks.
As a specific embodiment, optimal hyperlane can be utilized to carry out image block set described in recursive subdivision, thus form above-mentioned multistage tree construction.Fig. 4 shows the concrete example utilizing optimal hyperlane to carry out recursive subdivision image block set.Such as, breadth extreme segmentation rule (Maximum Spread Splitting Rule) used during kd can be adopted to set carrys out image block set described in recursive subdivision.Suppose that the image block set that will split represents with S, then split the dimension with breadth extreme that dimension can adopt this S set, the cut-point in this segmentation dimension then can adopt this S set along the median point of this segmentation dimension.As shown in Fig. 4 (A), by divisional plane P401, S set is divided into two classes, i.e. class 1 and class 2.Then, as shown in Fig. 4 (B), utilize divisional plane P402, class 1 is divided into again class 1-1 and class 1-2.As shown in Fig. 4 (C), utilize divisional plane P403, class 1-1 is divided into class 1-1-1 and class 1-1-2.As shown in Fig. 4 (D), utilize divisional plane P404, class 1-1-1 is divided into class 1-1-1-1 and class 1-1-1-2 again further.The rest may be inferred, adopts the mode of this recursive subdivision, and it is highly log that image block set S is divided into
2the binary tree structure of N, wherein N represents the number of the image block in image block set S.
Fig. 5 shows an example of the image block set being organized into multistage tree construction.Wherein, each root node or branch node can comprise the segmentation threshold information image block set of correspondence or subset being divided into two classes.
As a concrete example, each root node or branch node can comprise following information:
Cut_dim: the information of the dimension (namely split dimension) orthogonal with divisional plane;
Cut_val: the position (segmentation threshold value) of divisional plane;
Low_bnd: the image block set corresponding with this tree node or the lower limit of subset;
High_bnd: the image block set corresponding with this tree node or the upper limit of subset;
Left_child: to the link of the left child node of this node;
Right_child: to the link of the right child node of this node.
In addition, an element in each leaf node correspondence image set of blocks, i.e. an image block.Each leaf node can comprise following information:
Index: the index of the image block corresponding to this leaf node in image block set.
It should be understood, of course, that above-mentioned tree construction is only exemplary, and and be not intended to embodiment of the present disclosure to be limited to this.Any other tree constructions suitably can be adopted, do not enumerate here.
In the embodiment shown in Figure 2, because image block set have employed tree construction, therefore, when needing to inquire about this image block set in follow-up image processing process, first in tree construction, the leaf node with query object approximate match can be found fast according to the segmentation threshold information in each tree node.In other words, adopt such tree construction, traversal search can be carried out to whole tree construction, thus can greatly improve the speed of searching.Certainly, owing to splitting certain dimension in the divisional plane that adopts only representative image set of blocks space in the process of recursive subdivision image block set at every turn, the leaf node therefore searched at this tree construction is not likely the image block mated most with query object.But by the neighbour's collection searching for the leaf node found further, lookup result can be optimized further.
In some instances, any suitable method can be utilized to search for neighbour's image block of each image block.Such as, (as in step 103) can adopt the methods such as k nearest neighbor (K Nearest Neighbors) search procedure to search for neighbour's image block of certain image block, does not enumerate here.Fig. 3 provides the indicative flowchart searching for the method for neighbour's image block of certain image block in image block set according to a specific embodiment of the present disclosure.In the searching method of Fig. 3, one group of pseudo-point is linked with each image block, forms neighbour's collection of this each image block.
Suppose that image block P is an image block in image block set, as shown in Figure 3, the method (as step 103) of searching for neighbour's image block of this image block P in image block set can comprise step 103-2 to 103-7.In step 103-2, stochastic generation proper vector.The dimension of this proper vector is identical with the pixel count that image block P comprises, namely the two is in same feature space, and the distance of this proper vector and image block P is in predetermined scope, namely certain one dimension in image block set space of this proper vector or certain apteryx (can select the one dimension in image block set space or apteryx according to actual conditions; The dimension selected is more, then the neighbour's image block utilizing this proper vector to find is more accurate, is not construed as limiting here) to go up and the distance of image block is no more than certain threshold value.Should be understood that this predetermined scope or threshold value can set according to practical application scene, be not construed as limiting here.
In step 103-3, using generated proper vector as query object, in described image block set, search the image block nearest apart from this query object.Such as, any suitable searching method (as exhaustive search algorithm, exhaustive search) can be adopted to search for this arest neighbors image block at image block set, be not described further here.
In step 103-4, judge that whether the arest neighbors image block found is identical with this image block P, if not, then forward step 104-6 to; If so, then in step 103-5, using this arest neighbors image block as a pseudo-point, and the count value associated with this arest neighbors image block is increased; Particularly, if the arest neighbors image block that searches of epicycle with before a certain take turns or several arest neighbors image block obtained in search of taking turns identical, then its count value is increased one.If its count value with to search for the arest neighbors image block obtained before all not identical, is then set to one by the arest neighbors image block that epicycle searches.
In step 103-6, judge whether search has carried out T wheel, if so, then forwards step 103-7 to, otherwise, continue to repeat step 103-2 to 103-5, namely repeat described stochastic generation proper vector and described step of searching.Here T be greater than 1 integer, concrete value can set according to actual needs.That is, stochastic generation proper vector the step of searching the arest neighbors image block of this proper vector in image block set can perform repeatedly repeatedly, and the number of times of step that can determine to repeat stochastic generation proper vector according to the application scenarios of reality and search, generally speaking, the number of times repeated is more, the result obtained is more accurate, but calculated amount is also larger.Therefore, the value of T can be weighed according to actual needs, be not construed as limiting here.
In step 103-7, by find, one or more arest neighbors image block of associating with larger count value, the neighbour as this image block P collects.
In the method shown in Fig. 3, because the proper vector as query object of stochastic generation only has one dimension or apteryx, therefore, search for compared with neighbour's image block with adopting the image block of multidimensional as query object, can calculated amount be greatly reduced, thus improve the speed of search.
As a specific embodiment, the searching method shown in Fig. 3 can also use with methods combining image block set being organized into tree construction shown in Fig. 2.As mentioned above, not likely the image block mated most with query object at the leaf node utilizing the method shown in Fig. 2 to be organized into search in the image block set of tree construction.As shown in Figure 6, for the query point shown in figure, image block nearest with this query point in image block set is image block 2, instead of image block 1.But the divisional plane of reality may be P601, according to this divisional plane, in tree construction, inquire about the leaf node mated with this query point obtained is image block 1.By utilizing the method shown in Fig. 3, one group of pseudo-point can be linked with each leaf node (corresponding to an image block), forming neighbour's collection of each leaf node.Such as, neighbour's collection of image block 1 can comprise image block 2.Like this, concentrate search neighbour further, namely can obtain the image block mated the most with query point.
Particularly, for certain image block Pt (corresponding to the leaf node of in tree construction) in image block set, can first find in image block set space the unit Cell_Pt comprising this image block Pt (also referred to as bounding box, i.e. bounding box), as shown in Figure 7.This unit Cell-Pt corresponds in tree construction the minimum space unit split and obtain the multiple divisional planes corresponding to leaf node Pt and surround.Then, stochastic generation proper vector (being called query point) Qt in this unit Cell-Pt.Adopt such as exhaustive search algorithm in image block set, search the arest neighbors image block Pt_true of this query point.If the Pt_true found is different from image block Pt, then using this Pt_true as a pseudo-point, be placed in the pseudo-point list of image block Pt.If there is this Pt_true in pseudo-point list, then the frequency of occurrences count value of correspondence is added 1, otherwise, the frequency of occurrences count value of correspondence is set to 1.Repeatedly repeat above-mentioned stochastic generation query point to the step (concrete number of times can be determined according to practical application, be not construed as limiting) upgrading pseudo-point list here.Finally, according to corresponding frequency of occurrences count value by the sequence of pseudo-for each in pseudo-point list point, select one or more neighbour as the image block Pt collection occurring that frequency count is larger.Such as, the pseudo-point that frequency of occurrences count value in pseudo-point list is less than certain threshold value F_th can be removed.Threshold value F_th can set according to embody rule scene, is also not construed as limiting here.
Fig. 7 shows the schematic diagram of the puppet point utilizing the Cell_Pt corresponding with leaf node to find.Wherein, P705 and P706 represents actual divisional plane respectively, and the dark-shaded part that actual divisional plane surrounds is that Cell_Pt, P703 and P704 represent desirable divisional plane respectively.701 and 702 represent the zone errors that the needs puppet caused due to the error between actual divisional plane and desirable divisional plane is put, wherein, the region of the puppet point that the frequency of occurrences that 701 expressions find according to the method for above-mentioned example is high, the region of the pseudo-point (can remove) that the frequency of occurrences that 702 expressions find according to the method for above-mentioned example is low.
In this example, first find the unit Cell_Pt corresponding with leaf node Pt, therefore, can ensure that the proper vector (query point) of stochastic generation is all positioned at this unit, thus avoid the query point of generation error, ensure the accuracy of the neighbour's collection generated.In addition, by calculating the frequency of occurrences of each puppet point in pseudo-point list, can ensure only to retain the high pseudo-point of the frequency of occurrences, thus the speed of storage can be improved, and reducing required storage space.
The image processing equipment utilizing said method is embodiment still provides according to of the present disclosure.
Figure 10 shows the schematic block diagram of the image processing equipment according to an embodiment of the present disclosure.As shown in Figure 10, this image processing equipment 1000 can comprise image block set treating apparatus 1001 and information-storing device 1003.
Image block set treating apparatus 1001 can process the image block set comprising the multiple image blocks extracted from sample image.Particularly, image block set treating apparatus 1001 can search for one or more neighbour's image block of each image block in this image block set in described image block set.Similar to embodiment of the method/example above, neighbour's image block of each image block obtained by searching image set of blocks can be collected as the neighbour of each image block, and can comprise one or more nearest image block of each image block of this image block set middle distance.
What information-storing device 1003 can preserve that image block set and treating apparatus 1001 obtain collects relevant information with the neighbour of each image block in this set.
Utilizing the equipment shown in Figure 10, after the training set obtaining comprising multiple image block by training, neighbour's collection of each image block in this training set can be generated.While this training set of preservation, the information of neighbour's collection of wherein each image block can also be preserved.Like this, when needing to inquire about this training set in follow-up image procossing, after inquiry obtains certain image block, the neighbour that can also obtain this image block collects information, and can concentrate as required and this neighbour and carry out further search refinement, thus the image block more suited the requirements.With only generate and preserve the method for training and gathering or equipment is compared, the information that the equipment shown in Figure 10 generates more is conducive to follow-up inquiry, namely can inquire about obtaining information more accurately.
In certain embodiments, image block set can also be organized into the tree structure of classification by image block set treating apparatus 1001 further.Such as, can adopt and above with reference to the method described by figure 2, Fig. 4 and Fig. 5, image block set is organized into tree construction, no longer repeat here.
In certain embodiments, image block set treating apparatus 1001 can also adopt the neighbour's collection above carrying out each image block in searching image set of blocks with reference to the method described by figure 3, Fig. 6 and Fig. 7, also no longer repeats here.
In certain embodiments, the image block in the image block set in the above-mentioned and method/equipment that hereinafter will describe can be in groups, often organizes image block and comprises two or more image blocks corresponded to each other.Like this, the multiple neighbour's image blocks (as step 103 or 203) searching for each image block can comprise: as described in search is nearest with the image block often organized in image block in image block set one or more image block, the neighbour of a group or more groups image block corresponding for one or more image block described as this group image block is collected.In this case, if image block set is organized into tree construction, then each leaf node in this tree construction corresponds to one group of image block.Such as, in some Image Super Resolution Processing, the image block set comprising the multipair low resolution that corresponds to each other and high-definition picture block can be generated, namely a large amount of low resolution and high-definition picture sample can being utilized extracting the multipair low resolution that corresponds to each other and high-definition picture block, forming the training set comprising multipair image block.When the neighbour searching for each image block collects, only can search for neighbour's collection of one of often pair of image block, namely the image block of resolution corresponding to the image block that neighbour concentrates can be used as this and collects another neighbour in image block.In other words, often pair of image block neighbour's collection that can corresponding correspond to each other for a pair.When such as single low-resolution image generates high-definition picture in utilization, can using certain region in low-resolution image as query object, the low-resolution image block matched is inquired about in training set, thus obtain the high-definition picture block corresponding with this low-resolution image block, utilize this high-definition picture block both can rebuild the high-resolution areas corresponding with the region as query object.
Fig. 8 shows the indicative flowchart of the image processing method according to another embodiment of the present disclosure.In method shown in this Fig. 8, the image block set utilizing the method for example/embodiment above or equipment to obtain and neighbour collect information and carry out image procossing.
As shown in Figure 8, this image processing method can comprise step 813 and step 815.
In step 813, for certain query object, first can carry out coarse search in the image block set comprising the multiple image blocks extracted from sample image, obtain the image block approximate nearest with this query object distance.Should be understood that and any suitable searching method can be adopted to carry out this coarse search, such as, random search, Chaos Search or Equidistant search etc. can be adopted, do not enumerate here.
Each image block in image block set has neighbour's collection, and this neighbour collection comprises one or more nearest image block of described this image block of image block set middle distance.Therefore, after certain image block that coarse search arrives, search refinement can be carried out in step 815, namely concentrate its neighbour and search one or more image block nearest with described query object.Search the image block that obtains can as one or more arest neighbors image block of described query object by this refinement.
In the method shown in Fig. 8, the neighbour that make use of each image block in the image block set that method above or equipment generates collects information.By utilizing these information, inquiry can be divided into coarse search and search refinement two step, namely first pass through coarse search fast search to the result with query object approximate match, then carry out search refinement according to neighbour's collection of this result, thus find the result of mating the most.Like this, the speed of inquiring about can be accelerated, obtain Query Result accurately simultaneously.
In certain embodiments, image block set can be organized into hierarchical tree structure, wherein, and one or one group of image block in each leaf node correspondence image set of blocks.Root node can correspondence image set of blocks, and comprises the information of the thresholding for this set-partition being become the subset corresponding with next stage node.A subset of each branch node correspondence image set of blocks also comprises the information of the thresholding for the subset corresponding with this branch node being divided into the subset corresponding with next stage node.As a concrete example, image block set can be organized into structure as shown in Figure 5.In this case, the step (as step 813) of carrying out coarse search in described image block set can comprise: from as described in tree construction root node step by step in next stage with as described in the query object node motion of mating, until arrive the leaf node with described query object approximate match, wherein, in every one-level, select the node mated with described query object in next stage according to the information in node.Utilizing after this coarse search obtains the leaf node of certain and query object approximate match, being concentrated by the neighbour further at this leaf node and carrying out search refinement, the structure of mating the most with query object can be obtained.In the method, do not need to travel through whole image block set (tree construction), and the neighbour of leaf node concentrates the element comprised relatively less, therefore, obtaining, accurately while Query Result, the speed of inquiring about can also being accelerated.
Similar to some embodiment above, the image block in image block set can be in groups.Often organize image block and can comprise two or more image blocks corresponded to each other.
Such as, in some Image Super Resolution Processing, can generate the image block set comprising multipair image block, often pair of image block can comprise the low-resolution image block and high-definition picture block that correspond to each other for a pair.Particularly, a large amount of low resolution and high-definition picture sample can be utilized extracting the multipair low resolution that corresponds to each other and high-definition picture block, thus form the training set (or being called dictionary) comprising multipair image block.When utilizing low-resolution image to generate high-definition picture, such training set can be utilized.Fig. 9 shows and generates the schematic flow of the image super-resolution method of high-definition picture according to the low-resolution image that utilizes of a specific embodiment of the present disclosure.
As shown in Figure 9, the method can comprise step 913,915 and 919.
Step 913 is similar to step 813, is the step of carrying out coarse search at the image block set comprising multipair image block.Often pair of image block in image block set comprises the low-resolution image block and high-definition picture block that correspond to each other for a pair.These image blocks are all utilize a large amount of low resolution and high-definition picture sample to extracting.In addition, each image block in image block set has neighbour's collection.This neighbour collection can be adopt above described method or equipment to generate, and no longer repeats here.In order to utilize low-resolution image to rebuild high-definition picture, can using certain region in low-resolution image as query object.First, in image block set, carry out coarse search, obtain the image block with this query object approximate match.Similar to embodiment/example above, any suitable method can be adopted to carry out coarse search.Such as, when image block set is organized into tree construction, can step by step to the node motion of mating with described query object in next stage from the root node of tree construction, until arrive the leaf node with described query object approximate match, wherein, in every one-level, select the node mated with described query object in next stage according to the information in node.Additive method (as random search, Chaos Search or Equidistant search etc.) can also be adopted to carry out coarse search, no longer repeat here.
In step 915, coarse search to query object be similar to nearest image block neighbour concentrate carry out search refinement, to find one or more wherein nearest with query object image block, as one or more arest neighbors image block of this query object.These arest neighbors image blocks are all the image blocks of low resolution.
In step 919, first one or more high-definition picture block corresponding with one or more arest neighbors image block of query object is found, then, utilize these high-definition picture blocks to generate high-definition picture region, as part corresponding with described query object position in high-definition picture.Any suitable method can be adopted to generate described high-definition picture region to utilize one or more high-definition picture block, such as can adopt the correlation technique in following documents (being called pertinent literature 1): " Image Hallucination Using Neighbor EmbeddingOver Visual Primitive ManifoIds " (Proceedings of the IEEE ComputerSociety Conference0n Computer Vision and Pattern Recognition (CVPR) of the people such as Wei Fan, Minn., 18-23 day in June, 2007).The content of the document is incorporated herein by reference.
Like this, all above process is carried out to each region in low-resolution image, the high-definition picture corresponding with this low-resolution image can be rebuild.In method shown in Fig. 9, the neighbour that make use of each image block in the image block set that method above or equipment generates equally collects information.Like this, the speed of inquiring about can be accelerated, obtain Query Result accurately simultaneously.In addition, adopt the method for Fig. 9, can only utilize single width low-resolution image to generate high-definition picture.
Figure 11 shows the schematic block diagram of the image processing equipment according to an embodiment of the present disclosure.This image processing equipment 1100 is similar to the method shown in Fig. 8, make use of image block set that the method for example/embodiment above or equipment obtains and neighbour and collects information and carry out image procossing.
As shown in figure 11, image processing equipment 1100 can comprise coarse search device 1102 and search device for thinning 1104.
For certain query object, coarse search device 1102 can carry out coarse search in the image block set comprising the multiple image blocks extracted from sample image, obtains the image block approximate nearest with this query object distance.Should be understood that coarse search device 1102 can adopt any suitable searching method to carry out this coarse search, no longer repeat here.
Each image block in image block set has neighbour's collection, and this neighbour collection comprises one or more nearest image block of described this image block of image block set middle distance.Therefore, after certain image block that coarse search device 1102 searches, search device for thinning 1104 can be concentrated the neighbour of this image block and search one or more image block nearest with described query object, as one or more arest neighbors image block of described query object.
The neighbour of each image block in the image block set that the method above of the equipment utilization shown in Figure 11 or equipment generate collects information, can accelerate the speed of inquiring about, obtain Query Result accurately simultaneously.
Similar to some embodiments above, image block set can be organized into hierarchical tree structure, wherein, and one or one group of image block in each leaf node correspondence image set of blocks.Root node can correspondence image set of blocks, and comprises the information of the thresholding for this set-partition being become the subset corresponding with next stage node.A subset of each branch node correspondence image set of blocks also comprises the information of the thresholding for the subset corresponding with this branch node being divided into the subset corresponding with next stage node.As a concrete example, image block set can be organized into structure as shown in Figure 5.In this case, coarse search device 1102 can step by step to the node motion of mating with described query object in next stage from the root node of described tree construction, until arrive the leaf node with described query object approximate match, wherein, in every one-level, select the node mated with described query object in next stage according to the information in node.Namely the Output rusults of coarse search device 1102 is can be used as with the leaf node of query object approximate match.Do not need like this to travel through whole image block set (tree construction), and the neighbour of leaf node to collect the element comprised relatively less, therefore, obtaining, accurately while Query Result, the speed of inquiring about to be accelerated further.
Similar to some embodiment above, the image block in image block set can be in groups.Often organize image block and can comprise two or more image blocks corresponded to each other.
Figure 12 shows the schematic block diagram of the image processing equipment according to another specific embodiment of the present disclosure.This equipment 1200 is similar to equipment 1100, difference is, in this embodiment, image block in the image block set utilized is paired, often pair of image block comprises the low-resolution image block and high-definition picture block that correspond to each other for a pair, in addition, each image block in image block set has neighbour's collection.This neighbour collection can be adopt above described method or equipment to generate, and no longer repeats here.As shown in figure 12, except coarse search device 1202 and search device for thinning 1204, equipment 1200 also comprises resolution intensifier 1206.
Coarse search device 1202 using certain region in low-resolution image as query object, can carry out coarse search, obtains the image block with this query object approximate match in image block set.Similar to embodiment/example above, any suitable method can be adopted to carry out coarse search.Such as, when image block set is organized into tree construction, can step by step to the node motion of mating with described query object in next stage from the root node of tree construction, until arrive the leaf node with described query object approximate match, wherein, in every one-level, select the node mated with described query object in next stage according to the information in node.Additive method can also be adopted to carry out coarse search, no longer repeat here.
Search device for thinning 1204 device 1202 coarse search to query object be similar to nearest image block neighbour concentrate carry out search refinement, to find one or more wherein nearest with query object image block, as one or more arest neighbors image block of this query object.These arest neighbors image blocks are all the image blocks of low resolution.
Then, first resolution intensifier 1206 can find one or more high-definition picture block corresponding with one or more arest neighbors image block of query object, then, utilize these high-definition picture blocks to generate high-definition picture region, as part corresponding with described query object position in high-definition picture.Resolution intensifier 1206 can adopt any suitable method (method that the pertinent literature 1 such as above mentioned relates to) to utilize one or more high-definition picture block to generate described high-definition picture region, no longer repeats here.Like this, utilize each region in device 1202,1204 and 1206 pairs of low-resolution images all to carry out above process, the high-definition picture corresponding with this low-resolution image can be rebuild.
Similar to the embodiment of the method shown in Fig. 9, the neighbour that the equipment shown in Figure 12 make use of that method above or equipment generate, in image block set each image block equally collects information.Like this, the speed of inquiring about can be accelerated, obtain Query Result accurately simultaneously.In addition, the equipment of Figure 12 can only utilize single width low-resolution image to generate high-definition picture.
Figure 13 (a) shows a width low-resolution image.Figure 13 (b) shows the high-definition picture corresponding with low-resolution image that is Figure 13 (a) that be that utilize bilinear interpolation to generate.Figure 13 (c) shows the high-definition picture corresponding with the low-resolution image of Figure 13 (a) utilizing method or equipment with reference to the embodiment of the present disclosure shown in figure 9 or Figure 12 to generate.Figure 13 (d) shows the high-definition picture corresponding with the low-resolution image of Figure 13 (a) of true shooting.Can find out, the quality utilizing the high-definition picture generated according to method or the equipment of embodiment of the present disclosure is gratifying.
Should be understood that above-described embodiment and example are exemplary, instead of exhaustive, the disclosure should not be regarded as being confined to any specific embodiment or example.In embodiment of the present disclosure or example, involved image pattern or image block can be any forms, are not limited in any way here.
As an example, the software that each step of said method and all modules of the said equipment and/or device may be embodied as, firmware, hardware or its combination.Such as, can as the part of various equipment (as computing machine, personal digital assistant, camera, video camera and other image processing equipments etc., not enumerating here) with image processing function.When in the said equipment, all modules is configured by software, firmware, hardware or its mode combined, spendable concrete means or mode are well known to those skilled in the art, and do not repeat them here.
As mentioned above, the step processed image in said method and equipment and module can be realized by software.Install to the computing machine (multi-purpose computer 1400 such as shown in Figure 14) with specialized hardware structure the program forming this software from storage medium or network, this computing machine, when being provided with various program, can perform various functions etc.
In fig. 14, CPU (central processing unit) (CPU) 1401 performs various process according to the program stored in ROM (read-only memory) (ROM) 1402 or from the program that storage area 1408 is loaded into random access memory (RAM) 1403.In RAM1403, also store the data required when CPU1401 performs various process etc. as required.CPU1401, ROM1402 and RAM1403 are connected to each other via bus 1404.Input/output interface 1405 is also connected to bus 1404.
Following parts are connected to input/output interface 1405: importation 1406 (comprising keyboard, mouse etc.), output 1407 (comprise display, such as cathode-ray tube (CRT) (CRT), liquid crystal display (LCD) etc., and loudspeaker etc.), storage area 1408 (comprising hard disk etc.), communications portion 1409 (comprising network interface unit such as LAN card, modulator-demodular unit etc.).Communications portion 1409 is via network such as the Internet executive communication process.As required, driver 1410 also can be connected to input/output interface 1405.Detachable media 1411 such as disk, CD, magneto-optic disk, semiconductor memory etc. are installed on driver 1410 as required, and the computer program therefrom read is installed in storage area 1408 as required.
When series of processes above-mentioned by software simulating, from network such as the Internet or storage medium, such as detachable media 1411 installs the program forming software.
It will be understood by those of skill in the art that this storage medium is not limited to wherein having program stored therein shown in Figure 14, distributes the detachable media 1411 to provide program to user separately with equipment.The example of detachable media 1411 comprises disk (comprising floppy disk (registered trademark)), CD (comprising compact disc read-only memory (CD-ROM) and digital universal disc (DVD)), magneto-optic disk (comprising mini-disk (MD) (registered trademark)) and semiconductor memory.Or hard disk that storage medium can be ROM1402, comprise in storage area 1408 etc., wherein computer program stored, and user is distributed to together with comprising their equipment.
The disclosure also proposes a kind of program product storing the instruction code of machine-readable.When described instruction code is read by machine and performs, the above-mentioned method according to disclosure embodiment can be performed.
Correspondingly, be also included within of the present disclosure disclosing for carrying the above-mentioned storage medium storing the program product of the instruction code of machine-readable.Described storage medium includes but not limited to floppy disk, CD, magneto-optic disk, storage card, memory stick etc.
In the description above to disclosure specific embodiment, the feature described for a kind of embodiment and/or illustrate can use in one or more other embodiment by same or similar mode, combined with the feature in other embodiment, or substitute the feature in other embodiment.
Should emphasize, term " comprises/comprises " existence referring to feature, key element, step or assembly when using herein, but does not get rid of the existence or additional of one or more further feature, key element, step or assembly.
In addition, method of the present disclosure be not limited to specifications in describe time sequencing perform, also can according to other time sequencing ground, perform concurrently or independently.Therefore, the execution sequence of the method described in this instructions is not construed as limiting technical scope of the present disclosure.
Can be found out by above description, according to embodiment of the present disclosure, provide following scheme:
Remarks 1. 1 kinds of image processing methods, comprising:
In the image block set comprising the multiple image blocks extracted from sample image, search for one or more neighbour's image block of each image block, wherein, one or more neighbour's image block described comprises one or more nearest image block of this each image block of described image block set middle distance, thus forms neighbour's collection of this each image block; And
The information of preserving described image block set and collecting about the neighbour of each image block.
The method of remarks 2. as described in remarks 1, wherein, the multiple neighbour's image blocks searching for each image block comprise:
Stochastic generation proper vector, wherein the distance of this proper vector and this image block is in predetermined scope;
The arest neighbors image block nearest apart from described proper vector is searched in described image block set, if the arest neighbors image block found is different from this image block, then increase the count value of the arest neighbors image block association that this finds, wherein, stochastic generation at least two proper vectors search the arest neighbors image block of each proper vector; And
By find, one or more arest neighbors image block of associating with larger count value, the neighbour as this image block collects.
The method of remarks 3. as described in remarks 1, wherein, the image block in described image block set is in groups, and often group image block comprises two or more image blocks corresponded to each other, and the multiple neighbour's image blocks searching for each image block comprise:
One or more image block that search is nearest with the image block often organized in image block in described image block set, collects the neighbour of a group or more groups image block corresponding for one or more image block described as this group image block.
The method of remarks 4. as described in remarks 3, wherein, often organizes image block and comprises the low-resolution image block and high-definition picture block that correspond to each other for a pair.
The method of remarks 5. according to any one of remarks 1-4, also comprise: described image block set is organized into multistage tree construction, wherein, the corresponding described image block set of root node in described tree construction, and the information comprising the thresholding for described set-partition being become the subset corresponding with next stage node; A subset of the corresponding described image block set of each branch node also comprises the information of the thresholding for the subset corresponding with this branch node being divided into the subset corresponding with next stage node; And one group of image block in the corresponding described image block set of each leaf node.
The method of remarks 6. as described in remarks 5, wherein, the image block in described image block set is in groups, and often organizes image block and comprise two or more image blocks corresponded to each other, and described image block set is organized into multistage tree construction comprises:
Image block set described in recursive subdivision, wherein, is divided into two subsets by multiple series of images block to be split, until each subset comprises one group of image block in each segmentation.
Remarks 7. 1 kinds of image processing methods, comprising:
In the image block set comprising the multiple image blocks extracted from sample image, carry out coarse search, obtain the image block approximate nearest with query object distance; And
The neighbour of the image block arrived at coarse search concentrates, search one or more image block nearest with described query object, as one or more arest neighbors image block of described query object, wherein, coarse search to image block neighbour collection comprise one or more nearest image block of described this image block of image block set middle distance.
The method of remarks 8. as described in remarks 7, wherein, image block in described image block set is in groups, often organize image block and comprise the low-resolution image block and high-definition picture block that correspond to each other for a pair, and described query object is certain region in low-resolution image, and described method also comprises:
One or more high-definition picture block corresponding with one or more arest neighbors image block of described query object is utilized to generate high-definition picture region, as part corresponding with described query object position in high-definition picture.
The method of remarks 9. as described in remarks 7 or 8, wherein, described image block set is organized into hierarchical tree structure, and one or one group of image block in the corresponding described image block set of each leaf node of described tree construction, and in described image block set, carries out coarse search comprise:
Step by step to the node motion of mating with described query object in next stage from the root node of described tree construction, until arrive the leaf node mated with described query object, wherein, in every one-level, select the node mated with described query object in next stage according to the information in node.
Remarks 10. 1 kinds of image processing equipments, comprising:
Image block set treating apparatus, for in the image block set comprising the multiple image blocks extracted from sample image, search for one or more neighbour's image block of each image block, wherein, one or more neighbour's image block described comprises one or more nearest image block of this each image block of described image block set middle distance, thus forms neighbour's collection of this each image block; And
Information-storing device, for the information of preserving described image block set and collect about the neighbour of each image block.
The equipment of remarks 11. as described in remarks 10, wherein, described image block set treating apparatus is arranged to by the following multiple neighbour's image blocks searching for each image block:
Stochastic generation proper vector, wherein the distance of this proper vector and this image block is in predetermined scope;
The arest neighbors image block nearest apart from described proper vector is searched in described image block set, if the arest neighbors image block found is different from this image block, then increase the count value associated with the arest neighbors image block that this finds, wherein, stochastic generation at least two proper vectors search the arest neighbors image block of each proper vector; And
By find, one or more arest neighbors image block of associating with larger count value, the neighbour as this image block collects.
The equipment of remarks 12. as described in remarks 10, wherein, the image block in described image block set is in groups, and often group image block comprises two or more image blocks corresponded to each other, and described image block set treating apparatus is arranged to:
One or more image block that search is nearest with the image block often organized in image block in described image block set, collects the neighbour of a group or more groups image block corresponding for one or more image block described as this group image block.
The equipment of remarks 13. as described in remarks 12, wherein, often organizes image block and comprises the low-resolution image block and high-definition picture block that correspond to each other for a pair.
The equipment of remarks 14. according to any one of remarks 10-13, wherein, described image block set treating apparatus is also arranged to and described image block set is organized into multistage tree construction, wherein, the corresponding described image block set of root node in described tree construction, and the information comprising the thresholding for described set-partition being become the subset corresponding with next stage node; A subset of the corresponding described image block set of each branch node, and comprise the information of the thresholding for the subset corresponding with this branch node being divided into the subset corresponding with next stage node; And one group of image block in the corresponding described image block set of each leaf node.
The method of remarks 15. as described in remarks 14, wherein, image block in described image block set is in groups, and often organize image block and comprise two or more image blocks corresponded to each other, and described image block set treating apparatus is arranged to by following, described image block set is organized into multistage tree construction:
Image block set described in recursive subdivision, wherein, is divided into two subsets by multiple series of images block to be split, until each subset comprises one group of image block in each segmentation.
Remarks 16. 1 kinds of image processing equipments, comprising:
Coarse search device, for carrying out coarse search in the image block set comprising the multiple image blocks extracted from sample image, obtains the image block approximate nearest with query object distance; And
Search device for thinning, neighbour for the image block arrived at described coarse search device coarse search concentrates, search one or more image block nearest with described query object, as one or more arest neighbors image block of described query object, wherein, coarse search to image block neighbour collection comprise one or more nearest image block of described this image block of image block set middle distance.
The equipment of remarks 17. as described in remarks 16, wherein, image block in described image block set is in groups, often organize image block and comprise the low-resolution image block and high-definition picture block that correspond to each other for a pair, and described query object is certain region in low-resolution image, and described equipment also comprises:
Resolution intensifier, for utilizing one or more high-definition picture block corresponding with one or more arest neighbors image block of described query object to generate high-definition picture region, as part corresponding with described query object position in high-definition picture.
Although above by disclosing the disclosure the description of specific embodiment of the present disclosure, should be appreciated that, above-mentioned all embodiments and example are all illustrative, and not restrictive.Those skilled in the art can design various amendment of the present disclosure, improvement or equivalent in the spirit and scope of claims.These amendments, improvement or equivalent also should be believed to comprise in protection domain of the present disclosure.
Claims (6)
1. an image processing method, comprising:
In the image block set comprising the multiple image blocks extracted from sample image, search for one or more neighbour's image block of each image block, wherein, one or more neighbour's image block described comprises one or more nearest image block of this each image block of described image block set middle distance, thus forms neighbour's collection of this each image block; And
The information of preserving described image block set and collecting about the neighbour of each image block,
Wherein, the multiple neighbour's image blocks searching for each image block comprise:
Stochastic generation proper vector, wherein the distance of this proper vector and this image block is in predetermined scope;
The arest neighbors image block nearest apart from described proper vector is searched in described image block set, if the arest neighbors image block found is different from this image block, then increase the count value associated with the arest neighbors image block that this finds, wherein, stochastic generation at least two proper vectors search the arest neighbors image block of each proper vector; And
By find, one or more arest neighbors image block of associating with larger count value, the neighbour as this image block collects.
2. the image block the method for claim 1, wherein in described image block set is in groups, and often organizes image block and comprise two or more image blocks corresponded to each other, and the multiple neighbour's image blocks searching for each image block comprise:
One or more image block that search is nearest with the image block often organized in image block in described image block set, collects the neighbour of a group or more groups image block corresponding for one or more image block described as this group image block.
3. the method for claim 1, wherein, the image block in described image block set is in groups, often organizes image block and comprises the low-resolution image block and high-definition picture block that correspond to each other for a pair, and query object is certain region in low-resolution image, and described method also comprises:
One or more high-definition picture block corresponding with one or more arest neighbors image block of described query object is utilized to generate high-definition picture region, as part corresponding with described query object position in high-definition picture.
4. an image processing equipment, comprising:
Image block set treating apparatus, for in the image block set comprising the multiple image blocks extracted from sample image, search for one or more neighbour's image block of each image block, wherein, one or more neighbour's image block described comprises one or more nearest image block of this each image block of described image block set middle distance, thus forms neighbour's collection of this each image block; And
Information-storing device, for the information of preserving described image block set and collect about the neighbour of each image block,
Wherein, described image block set treating apparatus is arranged to by the following multiple neighbour's image blocks searching for each image block:
Stochastic generation proper vector, wherein the distance of this proper vector and this image block is in predetermined scope;
The arest neighbors image block nearest apart from described proper vector is searched in described image block set, if the arest neighbors image block found is different from this image block, then increase the count value associated with the arest neighbors image block that this finds, wherein, stochastic generation at least two proper vectors search the arest neighbors image block of each proper vector; And
By find, one or more arest neighbors image block of associating with larger count value, the neighbour as this image block collects.
5. equipment as claimed in claim 4, wherein, the image block in described image block set is in groups, and often group image block comprises two or more image blocks corresponded to each other, and described image block set treating apparatus is arranged to:
One or more image block that search is nearest with the image block often organized in image block in described image block set, collects the neighbour of a group or more groups image block corresponding for one or more image block described as this group image block.
6. equipment as claimed in claim 4, wherein, the image block in described image block set is in groups, often organizes image block and comprises the low-resolution image block and high-definition picture block that correspond to each other for a pair, and query object is certain region in low-resolution image, and described equipment also comprises:
Resolution intensifier, for utilizing one or more high-definition picture block corresponding with one or more arest neighbors image block of described query object to generate high-definition picture region, as part corresponding with described query object position in high-definition picture.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010257628.1A CN102375990B (en) | 2010-08-17 | 2010-08-17 | Method and equipment for processing images |
| JP2011178595A JP2012043437A (en) | 2010-08-17 | 2011-08-17 | Image processing method and image processing device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010257628.1A CN102375990B (en) | 2010-08-17 | 2010-08-17 | Method and equipment for processing images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102375990A CN102375990A (en) | 2012-03-14 |
| CN102375990B true CN102375990B (en) | 2015-03-04 |
Family
ID=45794568
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201010257628.1A Expired - Fee Related CN102375990B (en) | 2010-08-17 | 2010-08-17 | Method and equipment for processing images |
Country Status (2)
| Country | Link |
|---|---|
| JP (1) | JP2012043437A (en) |
| CN (1) | CN102375990B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103970769B (en) * | 2013-01-29 | 2018-06-26 | 华为技术有限公司 | Image search method and device |
| JP6278042B2 (en) | 2013-04-05 | 2018-02-14 | 日本電気株式会社 | Information processing apparatus and image processing method |
| CN103500176B (en) | 2013-09-06 | 2016-08-31 | 清华大学 | Sheet pessimistic concurrency control and construction method thereof |
| JP6800453B1 (en) * | 2020-05-07 | 2020-12-16 | 株式会社 情報システムエンジニアリング | Information processing device and information processing method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1378159A (en) * | 2001-03-26 | 2002-11-06 | Lg电子株式会社 | Picture searching method and device |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6226420B1 (en) * | 1997-12-26 | 2001-05-01 | International Business Machines Corporation | Non-integer scaling of raster images |
| US7474337B1 (en) * | 2000-10-24 | 2009-01-06 | Sony Corporation | Method and apparatus to provide edge enhancements as part of a demosaicing process |
| CN101536035B (en) * | 2006-08-31 | 2012-09-26 | 公立大学法人大阪府立大学 | Image recognition method, image recognition device, and image recognition program |
| CN101482925B (en) * | 2009-01-16 | 2012-01-04 | 西安电子科技大学 | Photograph generation method based on local embedding type hidden Markov model |
| CN101556601B (en) * | 2009-03-12 | 2011-08-10 | 华为技术有限公司 | Method and device for searching k neighbor |
-
2010
- 2010-08-17 CN CN201010257628.1A patent/CN102375990B/en not_active Expired - Fee Related
-
2011
- 2011-08-17 JP JP2011178595A patent/JP2012043437A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1378159A (en) * | 2001-03-26 | 2002-11-06 | Lg电子株式会社 | Picture searching method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| JP2012043437A (en) | 2012-03-01 |
| CN102375990A (en) | 2012-03-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112199375B (en) | Cross-modal data processing method and device, storage medium and electronic device | |
| CN110738207B (en) | Character detection method for fusing character area edge information in character image | |
| US10796244B2 (en) | Method and apparatus for labeling training samples | |
| US10366304B2 (en) | Localization and mapping method | |
| CN109886294A (en) | Knowledge fusion method, apparatus, computer equipment and storage medium | |
| CN102129560B (en) | Method and device for identifying characters | |
| CN104038792B (en) | For the video content analysis method and apparatus of IPTV supervision | |
| CN102375990B (en) | Method and equipment for processing images | |
| Jung et al. | Devil's on the edges: Selective quad attention for scene graph generation | |
| Xiao et al. | Deterministic model fitting by local-neighbor preservation and global-residual optimization | |
| CN114037055A (en) | Data processing system, method, device, equipment and storage medium | |
| Lee et al. | SSPNet: Learning spatiotemporal saliency prediction networks for visual tracking | |
| CN114996495A (en) | Single-sample image segmentation method and device based on multiple prototypes and iterative enhancement | |
| Fan et al. | A run-length-coding-based approach to stroke extraction of Chinese characters | |
| CN110135428A (en) | Image segmentation processing method and device | |
| CN113822134A (en) | Instance tracking method, device, equipment and storage medium based on video | |
| Sun et al. | Corner detection via scale-space behavior-guided trajectory tracing | |
| CN109885680B (en) | Short text classification preprocessing method, system and device based on semantic extension | |
| CN111428612A (en) | Pedestrian re-identification method, terminal, device and storage medium | |
| CN116340869A (en) | Distributed CatB body detection method and equipment based on red fox optimization algorithm | |
| CN113723515B (en) | Moire pattern recognition method, device, equipment and medium based on image recognition | |
| CN114638236A (en) | Intelligent question answering method, device, equipment and computer readable storage medium | |
| Montagner et al. | NILC: a two level learning algorithm with operator selection | |
| Jeune et al. | A unified framework for attention-based few-shot object detection | |
| JP6336827B2 (en) | Image search device, image search method, and search system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150304 Termination date: 20180817 |