CN106485266A - A kind of ancient wall classifying identification method based on extraction color characteristic - Google Patents
A kind of ancient wall classifying identification method based on extraction color characteristic Download PDFInfo
- Publication number
- CN106485266A CN106485266A CN201610846844.7A CN201610846844A CN106485266A CN 106485266 A CN106485266 A CN 106485266A CN 201610846844 A CN201610846844 A CN 201610846844A CN 106485266 A CN106485266 A CN 106485266A
- Authority
- CN
- China
- Prior art keywords
- image
- color
- characteristic
- characteristic vector
- point
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of based on the ancient wall classifying identification method for extracting color characteristic, it includes step 1, obtains the RGB color information of image, and is transformed into hsv color space；Step 2, the hsv color space of image is carried out unequal interval quantization；Step 3, statistics obtain the hsv color statistic histogram of image, used as the color feature vector of image；Step 4, normalization characteristic vector；Step 5, in image library, according to step 1～4 process each image, obtain the characteristic vector corresponding to each image；Step 6, for each class, extract such representative point using kmeans algorithm；Step 7, for each data point, calculate the characteristic vector of data point with each represent point characteristic vector similarity, and the corresponding image of data point is sorted out according to the class for representing point.The present invention improves the accuracy of ancient wall image recognition.
Description
Technical field
The invention belongs to computer pattern recognition, and in particular to a kind of identification of ancient wall image.
Background technology
Ancient wall identification is exactly that appliance computer mode identification technology is included into mural painting among affiliated classification, and then sentences
Break and drafting style, creation age and the affiliated region of fresco works.As ancient wall picture material is abundant, information is numerous
Many, therefore, accurately select needs the object of identification, and proposes corresponding solution for different problems." mural painting image
Packet many case-based learnings method in classification ", Tang great Wei etc., the 708-715 page, 2014 years of volume 19 of Journal of Image and Graphics
May describes a kind of packet many case-based learnings method of mural painting image, and the method proposes to divide according to the feature of mural painting image itself
Group policy, sample space is divided into different subspaces, per the independent train classification models of sub-spaces, is solved to a certain extent
The error in classification that sample is caused in the overall inseparable but situation of Locally separable.The accuracy of the method Classification and Identification is
82.21%.
Ph.D. Dissertation's " pictorial image sort research based on artistic style ", Yang Bing in June, 2013, the 1-113 page,
Propose a kind of pictorial image sorting technique based on artistic style similitude rule.Recognition methods for Dunhuang frescoes is root
Classified according to image inhomogeneity another characteristic, for example, mural painting is divided into shape, color, three kinds of different characteristics of illumination, and is entered with 50
Row is trained, and the result of image classification is：Constitute attribute 75%, color attribute 0.64%, illumination attribute 0.66%, finally all category
Property add up recognition correct rate be 78%.
Above first method proposes grouping strategy according to the feature of mural painting image itself, and recognition correct rate is 82.21%.
Shape, color, three kinds of methods of illumination are combined by second method, and recognition correct rate is 78%.It is in place of the deficiencies in the prior art：Know
Other accuracy is relatively low, it is necessary to improve recognition correct rate.
Technical term：
Data point：Refer to a pixel in piece image.
Represent a little：Refer to the most representative pixel extracted from each image.
Content of the invention
The technical problem to be solved is exactly to provide a kind of ancient wall classification based on extraction color characteristic to know
Other method, it can improve the accuracy of ancient wall image recognition.
The technical problem to be solved is realized by such technical scheme, and it comprises the following steps：
Step 1, the RGB color information of acquisition image, and it is transformed into hsv color space；
Step 2, the hsv color space of image is carried out unequal interval quantization；
Step 3, statistics obtain the hsv color statistic histogram of image, used as the color feature vector of image；
Step 4, normalization characteristic vector；
Step 5, in image library, according to step 1～4 process each image, obtain the characteristic vector corresponding to each image；
Step 6, for each class, extract such representative point using kmeans algorithm；
Step 7, for each data point, the characteristic vector for calculating data point represents the similar of the characteristic vector put with each
Degree, and the corresponding image of data point is sorted out according to the class for representing point.
Preferably, present invention additionally comprises：
Step 8, for each authentication image, according to fixed classification and the result of step 7 classification, just calculating classification
True rate.
The solution have the advantages that：Improve the accuracy of ancient wall image recognition.
Specific embodiment
The invention will be further described with reference to the accompanying drawings and examples：
The present invention is comprised the following steps：
Step 1, the RGB color information of acquisition image, and it is transformed into hsv color space；
As RGB color is a kind of color space related to hardware device, it is with human visual system to color
The subjective judgement of similitude does not meet, and is difficult to know the cognitive attribute of color represented by the value, therefore exists from RGB color
During graphical analysis, RGB color is generally transformed into the hsv color space for meeting human visual perception.
Step 2, the hsv color space of image is carried out unequal interval quantization；
Tri- components of H, S, V are carried out unequal interval quantization according to the color-aware of people, unequal interval quantify be exactly to face
The method that the division of each passage does not adopt decile in the colour space, but the size of its passage of man-made division.To color of image
Space carries out unequal interval and quantifies to make the metric form of color more meet cognition and judgement of the human eye to color.This step is adopted
Unequal interval quantification manner is that the chromatic component H in hsv color space is divided into 8 spaces along distribution axle, and saturation degree S is divided
For 3 spaces, it is H: S: V=8: 3: 3 that luminance component V is divided into 3 spaces, i.e. division proportion.I.e. tone H is divided into 8 parts, saturation degree
S and brightness V are each separated into 3 parts, the 3 color component synthesizing one-dimensional characteristic vectors for obtaining, according to tri- components of H, S, V in face
Color sensation gives different weights respectively on the significance level of visual signature impact in knowing：
G=9H+3S+V
In above formula, G represents that color characteristic quantifies collection, and H is expressed as 0～7 chrominance component；S be expressed as saturation degree component 0～
2；V is expressed as luminance component 0～2；So H, S, V, three color components are distributed on a n dimensional vector n and come, the span of G
For [0～71].
Step 3, statistics obtain the hsv color statistic histogram of image, used as the color feature vector of image；
For the color feature image using image, can be by the statistic histogram of color characteristic.Color of image is special
The statistic histogram that levies is really an one-dimensional discrete function, i.e.,：
In above formula, H is probability, and K is either element in G, and it is 0～71 that span is color feature value；nKIt is in image
With color feature value for the pixel of K number, N is the sum of image pixel.
After counting to G, image obtains the one-dimensional characteristic vector of 72 color components, and the one-dimensional characteristic is vectorial
Characteristic vector as image.
Step 4, normalization characteristic vector；
Normalization characteristic vector be each value in characteristic vector with divided by some value therein, typically divided by maximum
Value.
Step 5, in image library, according to step 1～4 process each image, obtain the characteristic vector corresponding to each image；
Step 6, for each class, extract such representative point using kmeans algorithm；
It is same class that the classification of image is the mural painting picture photographed in same region, and the picture that different regions photograph is
Inhomogeneity, for example：Dunhuang frescoes are a class, and Xinjiang mural painting is another kind of.
As the color of image feature of each class is less unified, therefore such face is represented using multiple color feature vectors
Color characteristic.For each class, clustered using kmeans algorithm, using the cluster point after cluster as class representative point.Therefore,
K value is selected to have a great impact for classifying quality in kmeans algorithm.
Kmeans algorithm is as follows：
K initial point of initial random selection, assigns to each cluster according to closest principle sample point to be sorted.Then by flat
All method recalculates the barycenter of each cluster, so that it is determined that the new cluster heart.Iteration always, until the displacement of the cluster heart is less than certain
Specified value, the representative point of the cluster heart as classification here.The number of cluster is set by the user.
So that 355 Guangyuans feel Jiange garden temple mural painting for classification 1 as an example, comprise the following steps：
1st, k image is selected at random as initial center in 355 mural painting images from classification 1；
2nd, remaining mural painting image distance respectively with initial center point is calculated, near just with that center with which initial center
Point is classified as a class (being used as the metric form of distance using Euclidean distance), until 355 mural painting images are divided into k by complete
Individual cluster；
3rd, after recalculating cluster by the method for mean value on the basis of point good class, the central point of cluster, repeats step 2；
4th, to the last algorithmic statement (i.e. the cluster heart no longer changes) then terminates, and obtaining input is：355 strokes in classification 1
The hsv color feature of image, a representative point quantity K cluster central point (k of as classification 1 is individual to be represented a little).
Kmeans algorithm is the function kmeans for directly invoking matlab, and the input of function kmeans is：In classification 1
Quantity k is put in the hsv color feature of 355 stroke pattern pictures and representative, then function kmeans is output as 1 (feel Jiange, Guangyuan of classification
Garden temple mural painting) k represent a little.
The representative point number of class is chosen after many experiments, and it is entirely different because of the difference of Mural painting content.
For example, according to the mural painting image data base that inventor is self-built, Guangyuan Jiange Jue Yuan temple mural painting in the mural painting image data base of Sichuan,
Guanghan Dragon Dwelling Temple mural painting, Xinjin Temple of Avalokitesvata mural painting, the class of newly numerous Long Zang temple mural painting and five class mural painting image of Pengxi Bao Fan temple mural painting
Represent point quantity to be respectively：4、2、6、2、3.Lijiang, yunnan white sand mural painting in national mural painting image data base, Beijing Fahai Temple
Mural painting, Shanxi Yongle Palace mural painting, Sichuan province mural painting, Kzier Reservir's mural painting, Mo kao grotto at Dunhuang mural painting, seven class figure of Tibet mural painting
The representative point quantity of picture is respectively：19、4、3、15、28、25、16.
Step 7, for each data point, the characteristic vector for calculating data point represents the similar of the characteristic vector put with each
Degree, and the corresponding image of data point is sorted out according to the class for representing point；
This step carries out the Similarity Measure of characteristic vector using histogram intersection method.Histogram intersection method is will be to be identified
Image zooming-out color histogram, calculates with the distance between each width color of image histogram in database that (each image is all
Known image in image library), distance image within the specific limits is and is judged as same category of image.
The process of histogram intersection method is：
H1And H (K)2(K) statistic histogram that two width color of image are characterized as K is respectively, then the matching value between two images
P can be realized by histogram intersection, i.e.,：
In above formula, P represents the histogram similarity of two width color of image features；Be color of image characteristic value from 0～
71 cumulative summation, min [H1(K), H2(K)] represent the minimum of a value taken in two values.H1And H (K)2(K) two width color of image are referred to
Feature value from G is the probability of K.The formula is defined asks for the intersecting minimum of a value of two width image histograms.
Classifying method according to histogram similarity P is：The color histogram of images to be recognized is extracted, by comparing two width
Minimum of a value during image is intersecting, calculates the distance between color of image histogram in images to be recognized and database, then will
Same category is included into apart from neighbouring image.
Each image is represented a little, then general character of each class image on color score is representative point, this step
Classified using k-nearest neighbor (k-Nearest Neighbor, abbreviation KNN), it is, according to closest one or
The classification of the several samples of person is determining the classification belonging to sample to be divided.
Step 8, for each authentication image, according to fixed classification and the result of step 7 classification, just calculating classification
True rate.
Authentication image is after the correct classification of known image, then is classified with the present invention, calculates the correct of the present invention
Rate.Classification accuracy is calculated includes three partial contents：1. correct the classification is calculated；2. the classification of mistake in computation；3. calculate correct
The percentage of discrimination.
Embodiment
The present invention is applied to the identification of Sichuan mural painting image, and for the mural painting image of Sichuan, colouring information is view picture figure
As most representational visual information, and multiple different color characters in same class mural painting, can be separated.Send out in this method
In bright, calculate data point and the similarity of point is represented, so as to data point is classified as the corresponding classification of closest representative point, implement
The automatic classification of mural painting image.
Using the test result of present method invention it is：The Average Accuracy of Sichuan mural painting picture system identification reaches
88.86952%, national mural painting image averaging recognition correct rate is 87.7160671%, is gradually promoted to 88% through many experiments
Left and right.
Compared with prior art, the present invention has the advantage that：Feature extraction by mural painting color, it is achieved that species
The automatic classification of the complicated mural painting image of various, style, and improve the accuracy of identification.
Claims (5)
1. a kind of based on the ancient wall classifying identification method for extracting color characteristic, it is characterized in that, comprise the following steps：
Step 1, the RGB color information of acquisition image, and it is transformed into hsv color space；
Step 2, the hsv color space of image is carried out unequal interval quantization；
Step 3, statistics obtain the hsv color statistic histogram of image, used as the color feature vector of image；
Step 4, normalization characteristic vector；
Step 5, in image library, according to step 1～4 process each image, obtain the characteristic vector corresponding to each image；
Step 6, for each class, extract such representative point using kmeans algorithm；
Step 7, the similarity of the characteristic vector put with each representative by each data point, the characteristic vector of calculating data point,
And the corresponding image of data point is sorted out according to the class for representing point.
2. according to claim 1 based on the ancient wall classifying identification method for extracting color characteristic, it is characterized in that, also wrap
Include：Step 8, for each authentication image, according to fixed classification and the result of step 7 classification, calculate the accuracy of classification.
3. according to claim 2 based on the ancient wall classifying identification method for extracting color characteristic, it is characterized in that, in step
In rapid 2, tri- components of H, S, V are carried out unequal interval quantization according to the color-aware of people, tone H is divided into 8 parts, saturation degree S
3 parts are each separated into brightness V, the 3 color component synthesizing one-dimensional characteristic vectors for obtaining, according to tri- components of H, S, V in face
Color sensation gives different weights respectively on the significance level of visual signature impact in knowing：
G=9H+3S+V
In above formula, G represents that color characteristic quantifies collection, and H is 0～7 chrominance component；S is saturation degree component 0～2；V is luminance component
0～2.
4. according to claim 3 based on the ancient wall classifying identification method for extracting color characteristic, it is characterized in that, in step
In rapid 3, the hsv color statistic histogram of described image is following discrete function：
In formula, H is probability, and K is either element in G, nKBe in image with color feature value for the pixel of K number, N is figure
Sum as pixel.
5. the ancient wall classifying identification method based on extraction color characteristic according to claim 4, is characterized in that, above-mentioned
In step 7, the characteristic vector of data point with the similarity of the characteristic vector of each representative point is：
In formula, P represents the histogram similarity of two width color of image features；Color of image characteristic value from 0～71 tired
Plus summation, min [H1(K), H2(K) minimum of a value taken in two values, H] are represented1And H (K)2(K) refer to that two width color of image features exist
In G, value is the probability of K.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610846844.7A CN106485266A (en) | 2016-09-23 | 2016-09-23 | A kind of ancient wall classifying identification method based on extraction color characteristic |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610846844.7A CN106485266A (en) | 2016-09-23 | 2016-09-23 | A kind of ancient wall classifying identification method based on extraction color characteristic |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106485266A true CN106485266A (en) | 2017-03-08 |
Family
ID=58268775
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610846844.7A Pending CN106485266A (en) | 2016-09-23 | 2016-09-23 | A kind of ancient wall classifying identification method based on extraction color characteristic |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106485266A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944553A (en) * | 2017-11-22 | 2018-04-20 | 浙江大华技术股份有限公司 | A kind of method for trimming and device of CNN models |
| CN108520546A (en) * | 2018-04-09 | 2018-09-11 | 武汉纺织大学 | A kind of method that efficient embroidery decorative pattern digital color feature is marked and redesigned |
| CN110647910A (en) * | 2019-08-12 | 2020-01-03 | 浙江浩腾电子科技股份有限公司 | Image similarity calculation method based on color quantization |
| CN111476253A (en) * | 2019-01-23 | 2020-07-31 | 阿里巴巴集团控股有限公司 | Clothing image classification method, clothing image classification device, clothing image classification method, clothing image classification device and clothing image classification equipment |
| US11651229B2 (en) | 2017-11-22 | 2023-05-16 | Zhejiang Dahua Technology Co., Ltd. | Methods and systems for face recognition |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104574960A (en) * | 2014-12-25 | 2015-04-29 | 宁波中国科学院信息技术应用研究院 | Traffic light recognition method |
| CN105069131A (en) * | 2015-08-14 | 2015-11-18 | 北京工业大学 | Capsule endoscopy image retrieval method based on visual vocabularies and local descriptors |
-
2016
- 2016-09-23 CN CN201610846844.7A patent/CN106485266A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104574960A (en) * | 2014-12-25 | 2015-04-29 | 宁波中国科学院信息技术应用研究院 | Traffic light recognition method |
| CN105069131A (en) * | 2015-08-14 | 2015-11-18 | 北京工业大学 | Capsule endoscopy image retrieval method based on visual vocabularies and local descriptors |
Non-Patent Citations (3)
| Title |
|---|
| 曹宁等: "基于视觉词袋模型的图像分类改进方法", 《电子设计工程》 * |
| 杨杰等: "基于颜色直方图的石材图像检索", 《武汉理工大学学报信息与管理工程版》 * |
| 雒伟群等: "基于SIFT的壁画图像相似性度量", 《科技创新与应用》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944553A (en) * | 2017-11-22 | 2018-04-20 | 浙江大华技术股份有限公司 | A kind of method for trimming and device of CNN models |
| US11651229B2 (en) | 2017-11-22 | 2023-05-16 | Zhejiang Dahua Technology Co., Ltd. | Methods and systems for face recognition |
| CN108520546A (en) * | 2018-04-09 | 2018-09-11 | 武汉纺织大学 | A kind of method that efficient embroidery decorative pattern digital color feature is marked and redesigned |
| CN111476253A (en) * | 2019-01-23 | 2020-07-31 | 阿里巴巴集团控股有限公司 | Clothing image classification method, clothing image classification device, clothing image classification method, clothing image classification device and clothing image classification equipment |
| CN111476253B (en) * | 2019-01-23 | 2024-04-02 | 阿里巴巴集团控股有限公司 | Clothing image classification method, device and equipment and image classification method and device |
| CN110647910A (en) * | 2019-08-12 | 2020-01-03 | 浙江浩腾电子科技股份有限公司 | Image similarity calculation method based on color quantization |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106485266A (en) | A kind of ancient wall classifying identification method based on extraction color characteristic | |
| CN104268593B (en) | The face identification method of many rarefaction representations under a kind of Small Sample Size | |
| CN107944431B (en) | A kind of intelligent identification Method based on motion change | |
| CN111738064B (en) | Haze concentration identification method for haze image | |
| CN109214298B (en) | Asian female color value scoring model method based on deep convolutional network | |
| CN103325122B (en) | Based on the pedestrian retrieval method of Bidirectional sort | |
| CN107066972B (en) | Natural scene Method for text detection based on multichannel extremal region | |
| CN109190643A (en) | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment | |
| CN106778852A (en) | A kind of picture material recognition methods for correcting erroneous judgement | |
| CN111143615B (en) | Short video emotion classification recognition device | |
| CN106897669A (en) | A kind of pedestrian based on consistent iteration various visual angles transfer learning discrimination method again | |
| CN109191428A (en) | Full-reference image quality evaluating method based on masking textural characteristics | |
| Casanova et al. | Texture analysis using fractal descriptors estimated by the mutual interference of color channels | |
| CN107305691A (en) | Foreground segmentation method and device based on images match | |
| CN105956570B (en) | Smiling face's recognition methods based on lip feature and deep learning | |
| CN110991389A (en) | Matching method for judging appearance of target pedestrian in non-overlapping camera view angle | |
| CN106529377A (en) | Age estimating method, age estimating device and age estimating system based on image | |
| CN104966075B (en) | A kind of face identification method and system differentiating feature based on two dimension | |
| CN109829924A (en) | A kind of image quality evaluating method based on body feature analysis | |
| CN106971158A (en) | A kind of pedestrian detection method based on CoLBP symbiosis feature Yu GSS features | |
| CN105303546B (en) | Neighbour's propagation clustering image partition method based on fuzzy connectedness | |
| CN109447111A (en) | A kind of remote sensing supervised classification method based on subclass training sample | |
| Warrell et al. | Labelfaces: Parsing facial features by multiclass labeling with an epitome prior | |
| CN114239754B (en) | Pedestrian attribute identification method and system based on attribute feature learning decoupling | |
| CN103049754B (en) | The picture recommendation method of social networks and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170308 |