CN101702239A - Method for describing texture of image with gradable wavelet packet transformation - Google Patents
Method for describing texture of image with gradable wavelet packet transformation Download PDFInfo
- Publication number
- CN101702239A CN101702239A CN200910024348A CN200910024348A CN101702239A CN 101702239 A CN101702239 A CN 101702239A CN 200910024348 A CN200910024348 A CN 200910024348A CN 200910024348 A CN200910024348 A CN 200910024348A CN 101702239 A CN101702239 A CN 101702239A
- Authority
- CN
- China
- Prior art keywords
- image
- gradable
- wavelet packet
- texture
- wavelet
- 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.)
- Pending
Links
Images
Abstract
The invention discloses a method for describing the texture of an image with gradable wavelet packet transformation, which is characterized by comprising the following implementing steps of; firstly, carrying out image gray processing on an input original image and transforming the image into a gray image; next, carrying out gradable wavelet packet transformation on the gray image; and finally carrying out texture characteristic description. In the process of texture characteristic description, the average value and the standard value of each subband are adopted, and finally texture describing characteristics of all subbands Wbll are merged to be used as the final testure description.
Description
Technical field
The present invention relates to the texture description method of image in the image analysis processing field, be specifically related to a kind of method for describing texture of image of the gradable wavelet package transforms based on local piecemeal.
Background technology
It is an important content in image/video analysis and the searching field that image is carried out effective feature description.Textural characteristics is the important description feature of in many features.In MPEG-7 standard and patent ZL.00804430.9, adopt the feature of Gabor transform domain to come image is carried out texture description.Gabor texture description symbol can effectively utilize the information under different angles and the change of scale subband to carry out texture description.But all these conversion are based on all that original image carries out, and therefore the textural characteristics of its description has certain limitation, and the computation complexity height of Gabor texture description symbol.
Summary of the invention
The present invention is that the gradable wavelet packet texture description symbol that is proposed is to propose at the weak point that Gabor texture description among MPEG-7 standard and the patent ZL.00804430.9 accords with, and its objective is in conjunction with the advantage of wavelet package transforms to come image is carried out effective texture description.
For reaching above purpose, the present invention takes following technical scheme to be achieved:
A kind of method for describing texture of image of gradable wavelet packet transformation is characterized in that, comprises following execution in step:
At first the original image to input carries out the image gray processing processing, and image transformation is become gray level image; Next gray level image is carried out gradable wavelet packet transformation; Carry out textural characteristics at last and describe, describe in the step, adopt the average and the standard deviation of each subband at textural characteristics, respectively shown in expression formula (1) and (2):
In the formula: W
Bl l(s is that (s, the coefficient of wavelet package transforms t), S and T are represented the height and the width of this sub-band images respectively in bl subband of l level t).S, t are illustrated respectively in the position on height and the Width, at last again with each subband W
Bl lThe texture description feature merge as final texture description.
In the said method, described image gray processing carries out the image local piecemeal earlier after handling, and the input gray level image is divided into identical four, five or nine of sizes, and the image to piecemeal carries out gradable wavelet packet transformation more then.Preferably divide five.Described gray level image is carried out in the gradable wavelet packet transformation step, adopt and decompose the wavelet packet tree method fully, and the decomposed class L in the wavelet package transforms is 4 grades.
Described with each subband W
Bl l(s, the method that texture description feature t) merges comprises four kinds of optional texture descriptions symbols, is respectively in the wavelet package transforms of expression afterbody corresponding 4
LThe average of individual subband and variance WVPK, the expression WVPK in except that the lowest frequency subband all 4
LThe average and the variance HWVP of all subbands under the average of-1 subband and variance NoDC, the average of representing high-frequency sub-band all under 0~L decomposition level and variance HIGH, the expression 0~L decomposition level.
Compared with prior art, the advantage of the inventive method is the texture description that gradable wavelet packet texture description method can be carried out different scale, different frequency bands to image.
Description of drawings
Fig. 1 is the scalable wavelet bag texture description method step block diagram of a kind of piecemeal among the present invention.
Fig. 2 is to original gray level image piecemeal not, is divided into four, five and nine synoptic diagram.
Fig. 3 is the division methods synoptic diagram of four kinds of dissimilar wavelet packet trees.
Fig. 4 is not next width of cloth original color image of piecemeal situation and through the design sketch of firsts and seconds wavelet package transforms of Fig. 2 (a).
Fig. 5 is the design sketch of next width of cloth original color image of five situations of Fig. 2 (c) and process firsts and seconds wavelet package transforms.
Embodiment
The present invention is described in further detail below in conjunction with drawings and Examples.
The invention provides a kind of scalable wavelet bag texture description of the overall situation or the wavelet packet texture description of several local piecemeals.In gradable wavelet packet texture description symbol, at first the image to monoblock (or each piecemeal) carries out the wavelet packet texture description, and then the texture description feature of each subband behind the conversion wavelet packet is merged as final texture description.Be that example is carried out the summary of the invention elaboration with overall wavelet packet texture description now.
In order to reach effective description to image texture characteristic, being implemented as follows based on scalable wavelet bag texture description symbol of being adopted among the present invention:
As shown in Figure 1, a kind of texture description method of gradable wavelet packet transformation of piecemeal comprises following execution in step:
At first the original image 00 of input being carried out gray processing handles 10 image transformation is become gray level image; Next, execution in step 20 is divided into localized mass with image; Execution in step 30 then, carry out gradable wavelet packet transformation; Last execution in step 40 is carried out textural characteristics and is described.
Wherein, image local piecemeal step 20 is the pieces that the input gray level image are divided into rule, each piece big or small identical.
Fig. 2 (a) to Fig. 2 (d) example provided, original gray level image is divided into one (not piecemeal), four, five and nine situation.Preferably use the branch block mode shown in Fig. 2 (c) among the present invention.
Gradable wavelet packet transformation step 30 is that each image block of being divided in the step 20 is carried out gradable wavelet package transforms.In gradable wavelet transformation, can adopt the division methods of the wavelet packet tree of rule, the division methods of the wavelet packet that also can define arbitrarily tree.
Fig. 3 (a) to Fig. 3 (d) example provided the division methods of four kinds of dissimilar wavelet packet trees.Wherein Fig. 3 (a) is the tree of the wavelet packet of decomposition fully under the decomposed class L=3.Fig. 3 (b) and Fig. 3 (c) are respectively two kinds of irregular decomposition wavelet packet trees.Fig. 3 (d) is wavelet decomposition ripple Bao Shu.The preferred L=4 that uses among the present invention, and adopt the described wavelet packet tree type that decomposes fully of Fig. 3 (a).Under above-mentioned preferable case, the title that to choose preferred employing wavelet filter be the DB1[wavelet filter of the structure of WAVELET PACKET DECOMPOSITION of the present invention and wavelet basis function also becomes the Haar wavelet filter, and this is to be known knowledge in this area.The length of its median filter is 2, and low pass wherein low-pass filter is [0.7071,0.7071], and Hi-pass filter is [0.7071,0.7071]].
Fig. 4 (a) to Fig. 4 (c) example provided piecemeal situation next width of cloth original color image not and through the figure as a result of firsts and seconds wavelet package transforms.Fig. 5 (a) to Fig. 5 (c) example provided and be divided into next width of cloth original color image of four situations and through the figure as a result of firsts and seconds wavelet package transforms.In Fig. 4 and Fig. 5, only provided the image wavelet packet transform result schematic diagram under the conversion progression L=2, the reason that does not show conversion progression L=4 hypograph wavelet package transforms result is because under the many situations of progression, corresponding subband is very many, and is less relatively by its resolution after the conversion.As can be seen from the figure,, the local grain information of image can be described effectively based on local block division method, thus can be to important help being arranged based on image similarity tolerance of texture information etc.
In the methods of the invention, to describe step 40 be that image block (comprising a not monoblock image of piecemeal) to each gradable wavelet packet transformation in the step 30 carries out texture description to textural characteristics.In textural characteristics is described, adopt the average μ of each subband
b lAnd standard deviation sigma
b l, respectively shown in expression formula (1) and (2)
W wherein
Bl l(s is that (s, the coefficient of wavelet package transforms t), S and T are represented the height and the width of this sub-band images respectively in bl subband of some l level t).S, t are illustrated respectively in the position on height and the Width.
Adopt four kinds of optional texture description symbols with each subband W among the present invention
Bl lThe texture description feature merges.It is respectively WVPK, NoDC, HIGH and HWVP.Wherein WVPK represents in the wavelet package transforms of afterbody corresponding 4
LThe average of individual subband and variance; NoDC represent among the WVPK except that the lowest frequency subband all 4
LThe average of-1 subband and variance; HIGH represents the average and the variance of high-frequency sub-band all under 0~L conversion stage; HWVP represents the average and the variance of subbands all under 0~L conversion stage.
Table 1 has provided different scalable wavelets and has wrapped in definition of texture description symbol and the textural characteristics dimension under different conversion progression under the block mode shown in Fig. 2 (a).Under the identical situation of scalable wavelet bag decomposed class, adopt four, the dimension of the scalable wavelet bag texture description symbol under five and the nine branch block modes is respectively 4,5 and 9 times that the corresponding texture description of institute accords with in the table 1.
The scalable wavelet bag texture description symbol definition that table 1. is different
The superiority based on the texture description method of the gradable wavelet packet transformation of local piecemeal that proposes in order to illustrate among the present invention has provided a kind of situation about this application of texture description symbol in image classification below.And carried out comparative analysis based on the Gabor texture description with existing.Suppose given texture description symbol be characterized as X, in the image classification to the K class, every class is chosen N training sample, the cluster centre computing method of then every class sample are as follows:
X wherein
i kBe the feature of a d dimension, K is the class number of object, X
i k(j) (j=1 ..., d) carry out normalized in the following way
Wherein I is an image collection.For a given test pattern, its corresponding texture description is characterized as X, and we are divided into k
0The method of class is according to the minimum criterion of distance
Dist (X, X wherein
k) feature X and X
kEuclidean distance, its computing method are as follows
We adopt 8 class images among the Spots Events (to please refer to paper L.Li in detail in experiment, and F.Li, and " What, where and who? Classifying events by scene and objectrecognition; " in Proc.ICCV, 2007.) analyze.All be fixed on 30 in every class number of training and also do test with remaining image, discovery is 21.6% based on the average correct recognition rata of the texture description method of Gabor, and the average correct recognition rata of the inventive method is 55.3%.Method of the present invention as can be seen is than existing Gabor texture description method excellence.
Through the above description of the embodiments, those skilled in the art can be well understood to the present invention and can realize by the mode that software adds essential hardware platform, can certainly all implement, but the former is better embodiment under a lot of situation by hardware.
Below only for the preferable embodiment of the present invention: as the type of the progression of preferred use wavelet package transforms, wavelet packet wave filter with divide block mode etc.; but protection scope of the present invention is not limited thereto; anyly be familiar with those skilled in the art in the technical scope that the present invention discloses; the variation that can expect easily or replacement all may fall within protection scope of the present invention.
Claims (5)
1. the method for describing texture of image of a gradable wavelet packet transformation is characterized in that, comprises following execution in step:
At first the original image to input carries out the image gray processing processing, and image transformation is become gray level image; Next gray level image is carried out gradable wavelet packet transformation; Carry out textural characteristics at last and describe, describe in the step, adopt the average and the standard deviation of each subband at textural characteristics, respectively shown in expression formula (1) and (2):
In the formula: W
Bl l(s is that (s, the coefficient of wavelet package transforms t), S and T are represented the height and the width of this sub-band images respectively in bl subband of l level t).S, t are illustrated respectively in the position on height and the Width, at last again with each subband W
Bl lThe texture description feature merge as final texture description.
2. the method for describing texture of image of gradable wavelet packet transformation as claimed in claim 1, it is characterized in that, described image gray processing carries out the image local piecemeal earlier after handling, the input gray level image is divided into identical four, five or nine of sizes, and the image to piecemeal carries out gradable wavelet packet transformation more then.
3. the method for describing texture of image of gradable wavelet packet transformation as claimed in claim 2 is characterized in that, described the input gray level image is divided into identical five of size.
4. the method for describing texture of image of gradable wavelet packet transformation as claimed in claim 1, it is characterized in that, described gray level image is carried out in the gradable wavelet packet transformation step, adopt and decompose the wavelet packet tree method fully, and the decomposed class L in the wavelet package transforms is 4 grades.
5. the method for describing texture of image of gradable wavelet packet transformation as claimed in claim 1 is characterized in that, and is described with each subband W
Bl lThe method that merges of texture description feature comprise four kinds of optional texture descriptions symbols, be respectively in the wavelet package transforms of expression afterbody corresponding 4
LThe average of individual subband and variance WVPK, the expression WVPK in except that the lowest frequency subband all 4
LThe average and the variance HWVP of all subbands under the average of-1 subband and variance NoDC, the average of representing high-frequency sub-band all under the O-L decomposition level and variance HIGH, the expression O-L decomposition level.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN200910024348A CN101702239A (en) | 2009-10-16 | 2009-10-16 | Method for describing texture of image with gradable wavelet packet transformation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN200910024348A CN101702239A (en) | 2009-10-16 | 2009-10-16 | Method for describing texture of image with gradable wavelet packet transformation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN101702239A true CN101702239A (en) | 2010-05-05 |
Family
ID=42157152
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN200910024348A Pending CN101702239A (en) | 2009-10-16 | 2009-10-16 | Method for describing texture of image with gradable wavelet packet transformation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101702239A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103324677A (en) * | 2013-05-24 | 2013-09-25 | 西安交通大学 | Hierarchical fast image global positioning system (GPS) position estimation method |
CN103971360A (en) * | 2013-02-05 | 2014-08-06 | 中国移动通信集团广东有限公司 | Method and device for calculating picture texture feature |
CN107862709A (en) * | 2017-09-28 | 2018-03-30 | 北京华航无线电测量研究所 | A kind of method for describing texture of image of multi-direction pattern concatenate rule |
-
2009
- 2009-10-16 CN CN200910024348A patent/CN101702239A/en active Pending
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103971360A (en) * | 2013-02-05 | 2014-08-06 | 中国移动通信集团广东有限公司 | Method and device for calculating picture texture feature |
CN103324677A (en) * | 2013-05-24 | 2013-09-25 | 西安交通大学 | Hierarchical fast image global positioning system (GPS) position estimation method |
CN103324677B (en) * | 2013-05-24 | 2017-02-01 | 西安交通大学 | Hierarchical fast image global positioning system (GPS) position estimation method |
CN107862709A (en) * | 2017-09-28 | 2018-03-30 | 北京华航无线电测量研究所 | A kind of method for describing texture of image of multi-direction pattern concatenate rule |
CN107862709B (en) * | 2017-09-28 | 2020-03-27 | 北京华航无线电测量研究所 | Image texture description method of multidirectional mode connection rule |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yao et al. | Wave-vit: Unifying wavelet and transformers for visual representation learning | |
US20160103900A1 (en) | Data structuring and searching methods and apparatus | |
Picard et al. | Improving image similarity with vectors of locally aggregated tensors | |
EP2126789B1 (en) | Improved image identification | |
US20130346412A1 (en) | System and method of detecting common patterns within unstructured data elements retrieved from big data sources | |
US20030236661A1 (en) | System and method for noise-robust feature extraction | |
He et al. | Rotation invariant texture descriptor using local shearlet-based energy histograms | |
EP2195765B1 (en) | Enhanced image identification | |
CN103052962A (en) | Rough wavelet granular space and classification of multispectral remote sensing image | |
CN100530231C (en) | Method for processing colour image | |
CN104850859A (en) | Multi-scale analysis based image feature bag constructing method | |
Yang et al. | On the stationarity of multivariate time series for correlation-based data analysis | |
Lotfi et al. | Combining wavelet transforms and neural networks for image classification | |
CN115471781A (en) | Intelligent garbage classification system and classification method thereof | |
CN101702239A (en) | Method for describing texture of image with gradable wavelet packet transformation | |
Nanni et al. | Set of texture descriptors for music genre classification | |
CN111079665A (en) | Morse code automatic identification method based on Bi-LSTM neural network | |
CN106503112A (en) | Video retrieval method and device | |
Pun | Rotation-invariant texture feature for image retrieval | |
Chang et al. | MS-SincResnet: Joint learning of 1D and 2D kernels using multi-scale SincNet and ResNet for music genre classification | |
Guan et al. | Deep learning with MCA-based instance selection and bootstrapping for imbalanced data classification | |
Pradhan et al. | Texture and color visual features based CBIR using 2D DT-CWT and histograms | |
US10191976B2 (en) | System and method of detecting common patterns within unstructured data elements retrieved from big data sources | |
Pun et al. | Rotation-invariant texture classification using a two-stage wavelet packet feature approach | |
Nazir et al. | Multi-view Gender Classification using hybrid Transformed |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20100505 |