CN103839074A - Image classification method based on matching of sketch line segment information and space pyramid - Google Patents
Image classification method based on matching of sketch line segment information and space pyramid Download PDFInfo
- Publication number
- CN103839074A CN103839074A CN201410062436.3A CN201410062436A CN103839074A CN 103839074 A CN103839074 A CN 103839074A CN 201410062436 A CN201410062436 A CN 201410062436A CN 103839074 A CN103839074 A CN 103839074A
- Authority
- CN
- China
- Prior art keywords
- image
- line segment
- sketch
- feature
- sketch line
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention belongs to the technical field of image processing, and particularly relates to an image classification method based on matching of sketch line segment information and a space pyramid. According to the implementation scheme, the method comprises the steps of 1) utilizing an initial sketch model to obtain an initial sketch image of an image, taking the direction of the sketch line segment of each sketch point as the direction of each sketch point, designing directional windows by taking the sketch points as the centers, enabling the area formed by overlaying the directional windows of all the sketch points to be a structural area, and enabling residual areas to be non-structural areas; 2) extracting multi-scale SIFT features from the structural area, and extracting single-scale SIFT features from the non-structural areas; 3) conducting clustering, projection and statistics on the SIFT features to obtain a feature F1; 4) obtaining a statistical feature F2 of the angle and the length of the sketch line segment; 5) conducting cascade connection on the F1 and the F2, and conducting classification. The method mainly solves the problems that due to the fact that the single-scale SIFT feature in the matching method of the space pyramid cannot express the detailed information of the image well, the classification accuracy is not high.
Description
Technical field
The invention belongs to technical field of image processing, relate to image classification method, can be used for the intelligent Classification Management of image and related application.Be specifically related to a kind of image classification method based on sketch line segment information and space pyramid coupling.
Background technology
Images Classification is the classical problem in computer vision field, apply very extensive, as video monitoring, medical diagnosis, image retrieval etc. all need to use image classification method.Along with the magnanimity of multi-medium data increases, the importance of Images Classification is more and more outstanding.Images Classification, conventionally according to the semantic content of image, such as specific scene, specific inclusion etc., adds different class labels to image, realizes Images Classification.The impact of the factor such as image is often subject to visual angle, illumination, block, has brought very large challenge to Images Classification.
Space pyramid matching process, i.e. Spatial Pyramid Matching(SPM), be a kind of image classification method of classics.SPM is that classical way " feature bag model " is Bag of Features(BOF) one expansion, it has added spatial information by introducing pyramid division on BOF method basis in BOF, and it comprises following five steps for Images Classification: 1. local feature extracts and describes; 2. build visual dictionary; 3. carry out proper vector quantification according to visual dictionary; 4. image is carried out to pyramid and divide the visual dictionary histogram that obtains subregion and calculate every sub regions, cascade forms last space pyramid feature; 5. training classifier is classified.The method is a kind of method based on local feature, because its simplicity and high efficiency enjoy high praise, uses very extensive aspect Images Classification and retrieval.
SPM sorting technique has higher classification accuracy, but, SPM extracts the stage by the intensive image block of adopting point methods image is divided into fixed size at local feature, and on image block, gradient direction is added up, obtain the SIFT(Scale Invariant Feature Transform of single scale) feature, but regional area importance concerning classification different in image is different, applicable local feature is also different, only use the intensive single scale SIFT feature detailed information of presentation video all sidedly of adopting a little, be unfavorable for classification.
Thereby for solve in the pyramid matching process of space single scale SIFT feature well the detailed information of presentation video cause the not high enough problem of classification accuracy, the present invention proposes a kind of image classification method based on sketch line segment information and space pyramid coupling, first with initial sketch model, image is divided into structural region and non-structural region, then press diverse ways in different regions and extract SIFT feature, then pass through cluster, projection and statistics are obtained the more effective description to whole image, combine with the statistical nature based on sketch line segment again, with presentation video better, carry out Images Classification.
Summary of the invention
The problem existing in order to solve prior art, the object of this invention is to provide a kind of image classification method based on sketch line segment information and space pyramid coupling, to improve Images Classification accuracy rate.
Realizing technical thought of the present invention is: utilizing primal sketch model is the initial sketch map that initial sketch model obtains image, according to the line segment information of initial sketch map, image is divided into structural region and non-structural region; Extract the single scale SIFT feature of different scale images piece at structural region, by Traditional Space pyramid matching process partitioned image piece and extract single scale SIFT feature, obtain one based on initial sketch line segment and space pyramidal feature by processing such as cluster, histogram projection, pyramid divisions at non-structural region; From the sketch line segment of image, obtain the statistical nature based on sketch line segment of a presentation video; Employing effective method carries out cascade then for classification by every width image based on the pyramidal feature of initial sketch line segment and space and the statistical nature based on sketch line segment.The method comprises the following steps:
Step 1, two image collections of given training image set and test pattern set, both common composition diagrams are as taxonomy database;
Step 2, be the primal sketch map of all images in initial sketch model extraction Images Classification database according to primal sketch model, be initial sketch map, on initial sketch map basis, further process, image is divided into structural region and non-structural region two parts;
Step 3, to the every piece image in Images Classification database, extracts the single scale SIFT(Scale Invariant Feature Transform of multiple yardstick image block at structural region) feature;
Step 4, to the every piece image in Images Classification database, at non-structural region according to space pyramid matching process partitioned image piece and extract single scale SIFT feature on image block;
Step 5, to the every piece image in Images Classification database, the single scale SIFT feature of the single scale SIFT feature of the multiple yardstick image block of the structural region of every piece image and non-structural region is put together, carry out cluster, the division of space pyramid, histogram projection by space pyramid matching process, obtain representing every piece image based on the pyramidal feature of initial sketch line segment and space, be designated as F_PSSPM;
Step 6 to the every piece image in Images Classification database, is extracted the statistical nature based on initial sketch line segment that represents this image from the line segment of its initial sketch map, is designated as F_PS;
Step 7, to the every piece image in Images Classification database, carries out cascade by F_PSSPM and F_PS by certain weight, obtains representing the feature F_WHOLE of this width image;
Step 8, by space pyramid coupling kernel function, using the feature F_WHOLE of every piece image in training set as training sample, trains and obtains sorter;
Step 9, will classify in sorter described in the feature F_WHOLE input step 8 of every piece image in test set, thereby obtain classification results;
Described in above-mentioned steps 2, image is divided into structural region and non-structural region two parts, after the initial sketch map of having extracted image, find every line segment in initial sketch map, design and the equidirectional direction window of its place sketch line segment on each sketch point of every sketch line segment, one side of direction window is parallel to sketch line segment, another side is perpendicular to sketch line segment, the size of direction window is (2n+1) × (2n+1), n is greater than 8 natural number, the direction window stack of all sketch points on every sketch line segment is obtained to areal map, in original image, be structural region with the region of areal map correspondence position, region in original image except structural region is non-structural region.
Described in above-mentioned steps 3, extract the single scale SIFT feature of multiple yardstick image block at structural region, different in the SIFT feature of its extraction and Traditional Space pyramid, be specially:
Original SIFT characteristic extraction procedure comprises five steps, build metric space, detect extreme point, accurately locate that extreme point, key point principal direction distribute and key point feature is described, wherein to describe this step be centered by key point, to get 16 × 16 fritter to be divided into 4 × 4 fritters to key point feature again, on each fritter, carry out statistics the cascade of gradient direction, obtain the SIFT feature of 128 dimensions;
Structural region is carried out to the division of multiple yardstick, obtain the image block (16 × 16,32 × 32,64 × 64) of multiple yardstick, and these image blocks are divided into 4 × 4 fritters again, in the time that being divided, structural region adopts the intensive point mode of adopting, 8 pixels of interblock step-length, the step-length of the piece as 16 × 16 is 8 pixels, the step-length of 64 × 64 piece is also 8 pixels, on each fritter, carry out statistics the cascade of gradient direction, obtain the single scale SIFT feature of 128 dimensions that represent these image blocks.
Obtaining described in above-mentioned steps 5 represent every piece image based on the pyramidal feature of initial sketch line segment and space, be specially:
5a) the single scale SIFT feature of the SIFT feature of the multiple yardstick image block of the structural region of training set image and non-structural region is put together, and therefrom randomly draw out m SIFT feature, m is given parameters, and general value is between 50000 to 200000;
5b) utilize k-means clustering algorithm to carry out cluster, obtain visual dictionary D=[d
l, d
2..., d
k], wherein, K represents the size of visual dictionary, i.e. the number of the cluster centre in k-means clustering algorithm, d
i(i=1,2 ..., K) and be a column vector, represent a vision word, i.e. cluster centre, the value of K is generally between 200 to several thousand;
5c) all SIFT features to the every piece image in Images Classification database, shine upon it to visual dictionary, by each SIFT Feature Mapping to cluster centre;
5d) every piece image being carried out to 3 layers of pyramid divides, three layers of pyramid divided and obtained respectively 1 × 1,2 × 2,4 × 4 totally 21 sub-blocks, in these sub-blocks, cluster classification under SIFT the feature corresponding image block in sub-block is carried out to statistics with histogram, in each sub-block, obtain a histogram, by 21 histogram cascades in sequence that obtain, obtain based on initial sketch model and the pyramidal feature F_PSSPM in space;
The statistical nature based on initial sketch line segment that extracts this image of expression from its initial sketch map line segment described in above-mentioned steps 6, is specially:
Every sketch line segment has an angle and length, angle is the angle of the x axle positive axis in sketch line segment and image coordinate system, be x axle positive axis in the direction of the clock around true origin rotate to this sketch line segment parallel process in the angle of rotating, length is the sketch point number on sketch line segment;
The angle that quantizes each sketch line segment in sketch map obtains angle character, quantize to 20 yardsticks by the even angle of every sketch line segment, be yardsticks of every 9 degree, 1～9 tolerance is turned to 1, 10～18 tolerance are turned to 2, 172～180 tolerance are turned to 20, wherein the angular range of sketch line segment is 0 to 180 degree, if the angle of certain sketch line segment is 180 degree, the value after its corresponding quantification is 20, every piece image in Images Classification database is carried out to pyramid division, on the each image-region obtaining, statistics drops on the line segment number on these 20 yardsticks, the corresponding histogram of each image-region, by all histograms according to given concatenated in order, obtain the statistical nature F_PS_1 based on sketch line segment angle of presentation video,
Equally, quantize each line segment length in sketch map and obtain sketch line segment length feature, quantize to 20 yardsticks by the length of every sketch line segment, the sketch line segment that is about to comprise 1 to 5 sketch point is quantified as 1, the line segment that comprises 6 to 10 sketch points is quantified as to 2, the line segment that is greater than 100 sketch points is quantified as to 20, wherein the length range of sketch line segment is 0 to sketch points up to a hundred, then every piece image in Images Classification database is carried out to pyramid division, on each image-region, statistics drops on the line segment number on these 20 yardsticks, the corresponding histogram of each image-region, by all histograms according to given concatenated in order, obtain the statistical nature F_PS_2 based on sketch line segment length of presentation video,
Finally from F_PS_1, F_PS_2, select and in conjunction with forming statistical nature F_PS.
Compared with prior art, the present invention has the following advantages:
1, the present invention introduced initial sketch model image is divided into structural region and non-structural region before extracting bottom local feature, extract the local feature of different scale in zones of different with the detailed information of presentation video better, and calculate thus based on the pyramidal feature of initial sketch line segment and space, presentation video better, makes classification more accurate;
2, according to the line segment information extraction in initial sketch map a statistical nature based on sketch line segment that can be used for presentation video, by its with based on the pyramidal feature cascade of initial sketch line segment and space, Description Image better, is conducive to classification.
Accompanying drawing explanation
Fig. 1 is realization flow figure of the present invention;
Fig. 2 is the example that region is divided;
Fig. 3 is the inventive method and the confusion matrix visual figure of traditional pyramid matching process on scene15 data set is all kinds of;
Embodiment
For making the object, technical solutions and advantages of the present invention clearer, referring to accompanying drawing, the present invention is described in more detail.
Fig. 1 is the process flow diagram of the image classification method of a kind of space pyramid coupling based on sketch line segment information of the present invention, and as shown in Figure 1, the method comprises the following steps:
Step 1, two image collections of given training image set and test pattern set, both common composition diagrams are as taxonomy database;
Step 2, be the primal sketch map of all images in initial sketch model extraction Images Classification database according to primal sketch model, be initial sketch map, on initial sketch map basis, further process, image is divided into structural region and non-structural region two parts;
Specific practice is, after the initial sketch map of having extracted image, find every line segment in initial sketch map, design and the equidirectional direction window of its place sketch line segment on each sketch point of every sketch line segment, one side of direction window is parallel to sketch line segment, another side is perpendicular to sketch line segment, the size of direction window is (2n+1) × (2n+1), n is greater than 8 natural number, the direction window stack of all sketch points obtains areal map, in original image, be structural region with the region of areal map correspondence position, region in original image except structural region is non-structural region,
In Fig. 2, provide image has been divided into structural region and the two-part example of non-structural region, Fig. 2 (a) is the former figure of the piece image in " MITstreet " classification in public data collection scene15, Fig. 2 (b) is the initial sketch map of former figure shown in Fig. 2 (a), square in Fig. 2 (c) is for centered by the sketch point that roundlet was comprised in scheming, the direction window of getting along this sketch point place sketch line segment direction, this window is parallel with sketch line segment on one side, vertical with sketch line segment on one side, get n=16, window size is 33 × 33, black part in Fig. 2 (d) is divided into the areal map obtaining, in original image with Fig. 2 (d) in the region of black part correspondence position be structural region, region in original image except structural region is non-structural region,
Step 3, to the every piece image in Images Classification database, extracts the single scale SIFT feature of multiple yardstick image block at structural region;
Original SIFT characteristic extraction procedure comprises five steps, build metric space, detect extreme point, accurately locate that extreme point, key point principal direction distribute and key point feature is described, wherein to describe this step be centered by key point, to get 16 × 16 fritter to be divided into 4 × 4 fritters to key point feature again, on each fritter, carry out statistics the cascade of gradient direction, obtain the SIFT feature of 128 dimensions;
And the SIFT using in Traditional Space pyramid matching process has omitted above four steps and obtain key point with intensive sampling, being divided into thick and fast overlapping step-length by image is 8 pixels, the image fritter of size 16 × 16, and using the center of image fritter as key point, directly carry out feature description, strict it has not possessed yardstick unchangeability in fact, and we are called single scale SIFT;
We carry out the division of multiple yardstick to structural region, obtain the image block (16 × 16,32 × 32,64 × 64) of multiple yardstick, and this image block is divided into 4 × 4 fritters again, in the time that being divided, structural region adopts the intensive point mode of adopting, interblock step-length is 8 pixels, the step-length of the piece as 16 × 16 is 8 pixels, the step-length of 64 × 64 piece is also 8 pixels, on each fritter, carry out statistics the cascade of gradient direction, obtain the single scale SIFT feature of 128 dimensions that represent this image block;
Step 4, to the every piece image in Images Classification database, at non-structural region according to Traditional Space pyramid matching process partitioned image piece and extract single scale SIFT feature on image block;
Be specially: in the intensive mode a little of adopting, image is divided into step-length as 8 pixels, the overlapping size fritter as 16 × 16, the single scale SIFT feature of calculating on these fritters at non-structural region according to Traditional Space pyramid matching process;
Step 5, to the every piece image in Images Classification database, the single scale SIFT feature of the single scale SIFT feature of the multiple yardstick image block of the structural region of every piece image and non-structural region is put together, carry out cluster, the division of space pyramid, histogram projection by Traditional Space pyramid matching process, obtain representing every piece image based on the pyramidal feature of initial sketch line segment and space, be designated as F_PSSPM;
Specific practice is:
5a) the single scale SIFT feature of the SIFT feature of the multiple yardstick image block of the structural region of training set image and non-structural region is put together, and therefrom randomly draw out m SIFT feature, m is given parameters, and general value is between 50000 to 200000;
5b) utilize k-means clustering algorithm to carry out cluster, obtain visual dictionary D=[d
l, d
2..., d
k], wherein, K represents the size of visual dictionary, i.e. the number of the cluster centre in k-means clustering algorithm, d
i(i=1,2 ..., K) and be a column vector, represent a vision word, i.e. cluster centre, the value of K is generally between 200 to several thousand;
5c) all SIFT features to the every piece image in Images Classification database, shine upon it to visual dictionary, by each SIFT Feature Mapping to cluster centre;
5d) every piece image being carried out to 3 layers of pyramid divides, three layers of pyramid divided and obtained respectively 1 × 1,2 × 2,4 × 4 totally 21 sub-blocks, in these sub-blocks, cluster classification under SIFT the feature corresponding image block in sub-block is carried out to statistics with histogram, in each sub-block, obtain a histogram, by 21 histogram cascades in sequence that obtain, obtain based on initial sketch model and the pyramidal feature F_PSSPM in space;
Step 6 to the every piece image in Images Classification database, is extracted the statistical nature based on initial sketch line segment that represents this image from the line segment of its initial sketch map, is designated as F_PS;
Every sketch line segment has an angle and length, angle is the angle of the x axle positive axis in sketch line segment and image coordinate system, be x axle positive axis in the direction of the clock around true origin rotate to this sketch line segment parallel process in the angle of rotating, length is the sketch point number on sketch line segment;
The angle that quantizes each sketch line segment in sketch map obtains angle character, by angle (its scope is 0 to 180 degree) uniform quantization to 20 yardstick of every sketch line segment, be yardsticks of every 9 degree, 1～9 tolerance is turned to 1, 10～18 tolerance are turned to 2, 172～180 tolerance are turned to 20, if the angle of certain sketch line segment is 180 degree, the value after its corresponding quantification is 20, every piece image in Images Classification database is carried out to pyramid division, on the each image-region obtaining, statistics drops on the line segment number on these 20 yardsticks, obtain a statistic histogram at each image-region, by all histograms according to given concatenated in order, obtain the statistical nature F_PS_1 based on sketch line segment angle of presentation video,
Equally, can quantize each line segment length in sketch map and obtain sketch line segment length feature, quantize to 20 yardsticks by the length of every sketch line segment (its scope be 0 to sketch points up to a hundred), the sketch line segment that is about to comprise 1 to 5 sketch point is quantified as 1, the line segment that comprises 6 to 10 sketch points is quantified as to 2, the line segment that is greater than 100 sketch points is quantified as to 20, then every piece image in Images Classification database is carried out to pyramid division, on each image-region, statistics drops on the line segment number on these 20 yardsticks, obtain a statistic histogram at each image-region, by all histograms according to given concatenated in order, obtain the statistical nature F_PS_2 based on sketch line segment length of presentation video,
Finally from F_PS_1, F_PS_2, select and in conjunction with forming statistical nature F_PS;
Step 7, to the every piece image in Images Classification database, carries out cascade by F_PSSPM and F_PS by certain weight, obtains representing the feature F_WHOLE of this width image;
Specific practice is: the thought that adopts cross validation, a part for abstract image taxonomy database is newly set up verification msg collection, on verification msg collection, by various weights, the F_PSSPM of every piece image and F_PS are carried out to cascade classification, determine last cascade weight according to classification accuracy;
Step 8, by space pyramid coupling kernel function, using the feature F_WHOLE of every piece image in training set as training sample, trains and obtains sorter;
Specific practice is: usage space pyramid coupling kernel function, and train a support vector machine (Support Vector Machine) as final sorter;
Step 9, will classify in sorter described in the feature F_WHOLE input step 8 of every piece image in test set, thereby obtain classification results.
Effect of the present invention can further illustrate by following emulation experiment:
In order to illustrate advantage of the present invention and feature, below this invention is applied in natural image classification and goes, obtain classification results.
(1) experimental situation
The hardware test platform of this experiment is: PC(Intel double-core 3.00GHz, 4G internal memory);
The software test platform of this experiment is: Windows764 bit manipulation system, vs2010, matlab2011b.
(2) experiment content
15 class natural scene images in public data collection scene15 are classified.The classification of this 15 class image is respectively " CALsuburb ", " MITcoast ", " MITforest ", " MIThighway ", " MITinsidecity ", " MITmountain ", " MITopencountry ", " MITstreet ", " MITtallbuilding ", " PARoffice ", " bedroom ", " industrial ", " kitchen ", " livingroom ", " store ", its corresponding picture number is: 241, 360, 328, 260, 308, 374, 410, 292, 356, 215, 216, 311, 210, 289, 315.
In experiment, each parameter is set to:
From each classification, choose at random the set of 100 image construction training images, remaining image construction test pattern set;
The corresponding parameter n of direction window while every piece image being carried out to region division with primal sketch model is set to 16, and direction window size is 33 × 33;
Need to be from extracting at random m SIFT feature, m value 100000 all SIFT features of training image set before obtaining visual dictionary by k-means cluster;
Cluster centre number K value when k-means cluster is 200;
By carrying out region division with primal sketch model, extract feature training classifier, the image in test pattern set is classified, obtain result as shown in table 1,
The classification accuracy of (SPM) method on scene-15 mated in table 1 the present invention with Traditional Space pyramid
| Method | Tradition SPM classification accuracy | This patent classification accuracy |
| Classification accuracy | 81.1% | 84.1% |
(SPM) method all kinds of classification accuracy on scene-15 is mated in table 2 the present invention with Traditional Space pyramid
| Item name | Tradition SPM classification accuracy | This patent classification accuracy |
| CALsuburb | 0.993 | 1.000 |
| MITcoast | 0.738 | 0.838 |
| MITforest | 0.934 | 0.943 |
| MIThighway | 0.888 | 0.875 |
| MITinsidecity | 0.889 | 0.837 |
| MITmountain | 0.876 | 0.894 |
| MITopencountry | 0.706 | 0.803 |
| MITstreet | 0.927 | 0.917 |
| MITtallbuilding | 0.906 | 0.930 |
| PARoffice | 0.913 | 0.904 |
| bedroom | 0.741 | 0.733 |
| industrial | 0.735 | 0.777 |
| kitchen | 0.636 | 0.736 |
| livingroom | 0.519 | 0.646 |
| store | 0.758 | 0.777 |
Fig. 3 is the inventive method and the classification confusion matrix visual figure of Traditional Space pyramid matching process on scene15 data set, Fig. 3 (a) is the classification confusion matrix visual figure of Traditional Space pyramid matching process on scene15 data set, and Fig. 3 (b) is the visual figure of the classification confusion matrix of the inventive method on scene15 data set.
From table 1, table 2 and Fig. 3, can find out, classification accuracy of the present invention is obviously better than Traditional Space pyramid matching image sorting technique.
Parts, technique and the english abbreviation that the present embodiment does not describe in detail belongs to well-known components or the conventional means of the industry, here not narration one by one.
Claims (5)
1. the image classification method based on sketch line segment information and space pyramid coupling, is characterized in that: the method comprises the following steps:
Step 1, two image collections of given training image set and test pattern set, both common composition diagrams are as taxonomy database;
Step 2, be the primal sketch map of all images in initial sketch model extraction Images Classification database according to primal sketch model, be initial sketch map, on initial sketch map basis, further process, image is divided into structural region and non-structural region two parts;
Step 3, to the every piece image in Images Classification database, extracts the single scale SIFT feature of multiple yardstick image block at structural region;
Step 4, to the every piece image in Images Classification database, at non-structural region according to space pyramid matching process partitioned image piece and extract single scale SIFT feature on image block;
Step 5, to the every piece image in Images Classification database, the single scale SIFT feature of the single scale SIFT feature of the multiple yardstick image block of the structural region of every piece image and non-structural region is put together, carry out cluster, the division of space pyramid, histogram projection by space pyramid matching process, obtain representing every piece image based on the pyramidal feature of initial sketch line segment and space, be designated as F_PSSPM;
Step 6 to the every piece image in Images Classification database, is extracted the statistical nature based on initial sketch line segment that represents this image from the line segment of its initial sketch map, is designated as F_PS;
Step 7, to the every piece image in Images Classification database, carries out cascade by F_PSSPM and F_PS by certain weight, obtains representing the feature F_WHOLE of this width image;
Step 8, by space pyramid coupling kernel function, using the feature F_WHOLE of every piece image in training set as training sample, trains and obtains sorter;
Step 9, will classify in sorter described in the feature F_WHOLE input step 8 of every piece image in test set, thereby obtain classification results.
2. a kind of image classification method based on sketch line segment information and space pyramid coupling as claimed in claim 1, it is characterized in that: described in step 2, image is divided into structural region and non-structural region two parts, after the initial sketch map of having extracted image, find every line segment in initial sketch map, design and the equidirectional direction window of its place sketch line segment on each sketch point of every sketch line segment, one side of direction window is parallel to sketch line segment, another side is perpendicular to sketch line segment, the size of direction window is (2n+1) × (2n+1), n is greater than 8 natural number, the direction window stack of all sketch points on every sketch line segment is obtained to areal map, in original image, be structural region with the region of areal map correspondence position, region in original image except structural region is non-structural region.
3. a kind of image classification method based on sketch line segment information and space pyramid coupling as claimed in claim 1, it is characterized in that: described in step 3, extract the single scale SIFT feature of multiple yardstick image block at structural region, different in the SIFT feature of its extraction and Traditional Space pyramid, be specially:
Original SIFT characteristic extraction procedure comprises five steps, build metric space, detect extreme point, accurately locate that extreme point, key point principal direction distribute and key point feature is described, wherein to describe this step be centered by key point, to get 16 × 16 fritter to be divided into 4 × 4 fritters to key point feature again, on each fritter, carry out statistics the cascade of gradient direction, obtain the SIFT feature of 128 dimensions;
Structural region is carried out to the division of multiple yardstick, obtain the image block (16 × 16,32 × 32,64 × 64) of multiple yardstick, and these image blocks are divided into 4 × 4 fritters again, in the time that being divided, structural region adopts the intensive point mode of adopting, 8 pixels of interblock step-length, the step-length of the piece as 16 × 16 is 8 pixels, the step-length of 64 × 64 piece is also 8 pixels, on each fritter, carry out statistics the cascade of gradient direction, obtain the single scale SIFT feature of 128 dimensions that represent these image blocks.
4. a kind of image classification method based on sketch line segment information and space pyramid coupling as claimed in claim 1, is characterized in that: obtaining described in step 5 represent every piece image based on the pyramidal feature of initial sketch line segment and space, be specially:
5a) the single scale SIFT feature of the SIFT feature of the multiple yardstick image block of the structural region of training set image and non-structural region is put together, and therefrom randomly draw out m SIFT feature, m is given parameters, and value is between 50000 to 200000;
5b) utilize k-means clustering algorithm to carry out cluster, obtain visual dictionary D=[d
l, d
2..., d
k], wherein, K represents the size of visual dictionary, i.e. the number of the cluster centre in k-means clustering algorithm, d
i(i=1,2 ..., K) and be a column vector, represent a vision word, i.e. cluster centre, the value of K is generally between 200 to several thousand;
5c) all SIFT features to the every piece image in Images Classification database, shine upon it to visual dictionary, by each SIFT Feature Mapping to cluster centre;
5d) every piece image being carried out to 3 layers of pyramid divides, three layers of pyramid divided and obtained respectively 1 × 1,2 × 2,4 × 4 totally 21 sub-blocks, in these sub-blocks, cluster classification under SIFT the feature corresponding image block in sub-block is carried out to statistics with histogram, in each sub-block, obtain a histogram, by 21 histogram cascades in sequence that obtain, obtain based on initial sketch model and the pyramidal feature F_PSSPM in space.
5. a kind of image classification method based on sketch line segment information and space pyramid coupling as claimed in claim 1, it is characterized in that: the statistical nature based on initial sketch line segment that extracts this image of expression from its initial sketch map line segment described in step 6, is specially:
Every sketch line segment has an angle and length, angle is the angle of the x axle positive axis in sketch line segment and image coordinate system, be x axle positive axis in the direction of the clock around true origin rotate to this sketch line segment parallel process in the angle of rotating, length is the sketch point number on sketch line segment;
The angle that quantizes each sketch line segment in sketch map obtains angle character, quantize to 20 yardsticks by the even angle of every sketch line segment, be yardsticks of every 9 degree, 1～9 tolerance is turned to 1, 10～18 tolerance are turned to 2, 172～180 tolerance are turned to 20, wherein the angular range of sketch line segment is 0 to 180 degree, if the angle of certain sketch line segment is 180 degree, the value after its corresponding quantification is 20, every piece image in Images Classification database is carried out to pyramid division, on the each image-region obtaining, statistics drops on the line segment number on these 20 yardsticks, the corresponding histogram of each image-region, by all histograms according to given concatenated in order, obtain the statistical nature F_PS_1 based on sketch line segment angle of presentation video,
Equally, quantize each line segment length in sketch map and obtain sketch line segment length feature, quantize to 20 yardsticks by the length of every sketch line segment, the sketch line segment that is about to comprise 1 to 5 sketch point is quantified as 1, the line segment that comprises 6 to 10 sketch points is quantified as to 2, the line segment that is greater than 100 sketch points is quantified as to 20, wherein the length range of sketch line segment is 0 to sketch points up to a hundred, then every piece image in Images Classification database is carried out to pyramid division, on each image-region, statistics drops on the line segment number on these 20 yardsticks, the corresponding histogram of each image-region, by all histograms according to given concatenated in order, obtain the statistical nature F_PS_2 based on sketch line segment length of presentation video,
Finally from F_PS_1, F_PS_2, select and in conjunction with forming statistical nature F_PS.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410062436.3A CN103839074B (en) | 2014-02-24 | 2014-02-24 | Image classification method based on matching of sketch line segment information and space pyramid |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410062436.3A CN103839074B (en) | 2014-02-24 | 2014-02-24 | Image classification method based on matching of sketch line segment information and space pyramid |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103839074A true CN103839074A (en) | 2014-06-04 |
| CN103839074B CN103839074B (en) | 2017-02-08 |
Family
ID=50802552
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410062436.3A Active CN103839074B (en) | 2014-02-24 | 2014-02-24 | Image classification method based on matching of sketch line segment information and space pyramid |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103839074B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447488A (en) * | 2015-12-15 | 2016-03-30 | 西安电子科技大学 | SAR (synthetic aperture radar) image target detection method based on sketch line segment topological structure |
| CN105654122A (en) * | 2015-12-28 | 2016-06-08 | 江南大学 | Spatial pyramid object identification method based on kernel function matching |
| CN107545216A (en) * | 2016-06-29 | 2018-01-05 | 深圳市格视智能科技有限公司 | A kind of vehicle identification method available for power-line patrolling |
| CN108846843A (en) * | 2018-04-03 | 2018-11-20 | 南昌奇眸科技有限公司 | A kind of image characteristic extracting method |
| CN109409388A (en) * | 2018-11-07 | 2019-03-01 | 安徽师范大学 | A kind of bimodulus deep learning based on graphic primitive describes sub- building method |
| CN110334776A (en) * | 2019-07-15 | 2019-10-15 | 哈尔滨理工大学 | A kind of image classification recognition methods based on region bicubic interpolation technology |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101315663A (en) * | 2008-06-25 | 2008-12-03 | 中国人民解放军国防科学技术大学 | Nature scene image classification method based on area dormant semantic characteristic |
| CN102208038A (en) * | 2011-06-27 | 2011-10-05 | 清华大学 | Image classification method based on visual dictionary |
| GB2485573A (en) * | 2010-11-19 | 2012-05-23 | Alan Geoffrey Rainer | Identifying a Selected Region of Interest in Video Images, and providing Additional Information Relating to the Region of Interest |
-
2014
- 2014-02-24 CN CN201410062436.3A patent/CN103839074B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101315663A (en) * | 2008-06-25 | 2008-12-03 | 中国人民解放军国防科学技术大学 | Nature scene image classification method based on area dormant semantic characteristic |
| GB2485573A (en) * | 2010-11-19 | 2012-05-23 | Alan Geoffrey Rainer | Identifying a Selected Region of Interest in Video Images, and providing Additional Information Relating to the Region of Interest |
| CN102208038A (en) * | 2011-06-27 | 2011-10-05 | 清华大学 | Image classification method based on visual dictionary |
Non-Patent Citations (2)
| Title |
|---|
| 李梦雄: ""基于视觉计算和混合尺度局部特征的图像分类方法"", 《中国优秀硕士学位论文全文数据库信息科技辑 》 * |
| 郝红侠 等: ""采用结构自适应窗的非局部均值图像去噪算法"", 《西安交通大学学报》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447488A (en) * | 2015-12-15 | 2016-03-30 | 西安电子科技大学 | SAR (synthetic aperture radar) image target detection method based on sketch line segment topological structure |
| CN105447488B (en) * | 2015-12-15 | 2021-08-20 | 西安电子科技大学 | SAR image target detection method based on sketch line segment topological structure |
| CN105654122A (en) * | 2015-12-28 | 2016-06-08 | 江南大学 | Spatial pyramid object identification method based on kernel function matching |
| CN107545216A (en) * | 2016-06-29 | 2018-01-05 | 深圳市格视智能科技有限公司 | A kind of vehicle identification method available for power-line patrolling |
| CN108846843A (en) * | 2018-04-03 | 2018-11-20 | 南昌奇眸科技有限公司 | A kind of image characteristic extracting method |
| CN109409388A (en) * | 2018-11-07 | 2019-03-01 | 安徽师范大学 | A kind of bimodulus deep learning based on graphic primitive describes sub- building method |
| CN109409388B (en) * | 2018-11-07 | 2021-08-27 | 安徽师范大学 | Dual-mode deep learning descriptor construction method based on graphic primitives |
| CN110334776A (en) * | 2019-07-15 | 2019-10-15 | 哈尔滨理工大学 | A kind of image classification recognition methods based on region bicubic interpolation technology |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103839074B (en) | 2017-02-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108108751B (en) | Scene recognition method based on convolution multi-feature and deep random forest | |
| CN103839074A (en) | Image classification method based on matching of sketch line segment information and space pyramid | |
| CN106776856B (en) | Vehicle image retrieval method integrating color features and vocabulary tree | |
| CN102693299B (en) | System and method for parallel video copy detection | |
| CN106845341B (en) | Unlicensed vehicle identification method based on virtual number plate | |
| CN105005786A (en) | Texture image classification method based on BoF and multi-feature fusion | |
| Tabia et al. | Compact vectors of locally aggregated tensors for 3D shape retrieval | |
| Zeng et al. | Curvature bag of words model for shape recognition | |
| CN102663401B (en) | Image characteristic extracting and describing method | |
| CN101986295B (en) | Image clustering method based on manifold sparse coding | |
| CN104199842A (en) | Similar image retrieval method based on local feature neighborhood information | |
| CN112528934A (en) | Improved YOLOv3 traffic sign detection method based on multi-scale feature layer | |
| CN103440501A (en) | Scene classification method based on nonparametric space judgment hidden Dirichlet model | |
| CN104850859A (en) | Multi-scale analysis based image feature bag constructing method | |
| CN108764302A (en) | A kind of bill images sorting technique based on color characteristic and bag of words feature | |
| CN105930873A (en) | Self-paced cross-modal matching method based on subspace | |
| CN106203448B (en) | A kind of scene classification method based on Nonlinear Scale Space Theory | |
| CN103473308B (en) | High-dimensional multimedia data classifying method based on maximum margin tensor study | |
| CN105654122A (en) | Spatial pyramid object identification method based on kernel function matching | |
| Wang et al. | Separable vocabulary and feature fusion for image retrieval based on sparse representation | |
| CN108763266B (en) | Trademark retrieval method based on image feature extraction | |
| CN107045520B (en) | Vehicle image retrieval method based on position information weighted vocabulary tree | |
| CN111414958B (en) | Multi-feature image classification method and system for visual word bag pyramid | |
| CN104008095A (en) | Object recognition method based on semantic feature extraction and matching | |
| Karmakar et al. | An enhancement to the spatial pyramid matching for image classification and retrieval |
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 |