CN103226711A - Quick Haar wavelet feature object detecting method - Google Patents
Quick Haar wavelet feature object detecting method Download PDFInfo
- Publication number
- CN103226711A CN103226711A CN2013101047392A CN201310104739A CN103226711A CN 103226711 A CN103226711 A CN 103226711A CN 2013101047392 A CN2013101047392 A CN 2013101047392A CN 201310104739 A CN201310104739 A CN 201310104739A CN 103226711 A CN103226711 A CN 103226711A
- Authority
- CN
- China
- Prior art keywords
- quick
- haar
- detection method
- object detection
- haar 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a quick Haar wavelet feature object detecting method. According to the quick Haar wavelet feature object detecting method, through reasonable division of functional blocks and application of not more than 100-dimensional rectangular features, optimization of an integral image is achieved, the data processing speed is enhanced, and quick detection on an object to be detected is achieved.
Description
Technical field
The present invention relates to a kind of image detecting technique, particularly relate to a kind of being applicable to and in computer vision and man-machine interaction, use the application technology of Haar wavelet character optimization object detection.
Background technology
The Haar feature is used the most extensively in the wavelet transformation (Wavelet transform), and in the current computer vision was used, haar feature calculation speed was fast, but it is not fully up to expectations to detect aspect such as consuming time.And be subjected to the storage space resource limit, and can not use cpu resource such as similar PC at embedded platforms such as appliance systems.Make data processing method more fast so be necessary research, increase the adaptability of disposal route, and then the application method is had higher requirement.
In the computer vision industry, the more and more data disposal route is arisen at the historic moment for promoting processing speed.Class Haar(Haar-like) wavelet character enters the scope of commercial product application, and emphasis is optimized on the data processing method aspect.By class Haar wavelet character object moving detection fast, application process can be realized the use to limited resources, and the processing speed that obtains being exceedingly fast, and satisfies requirements such as subsequent image processing and recognition methods.
Summary of the invention
The technical problem to be solved in the present invention provides a kind of use Fast-Haar like and realizes quick object detection method, and this method has been accelerated data processing speed by the optimization process to integral image, realizes the fast detecting to object to be measured.
The technical scheme that the present invention or invention are adopted is as follows:
A kind of quick Haar wavelet character object detection method is characterized in that:
Step 1, use Haar-like feature are optimized integral image, realize the fast detecting to object;
Step 2, use are carried out dimensionality reduction based on the cascade classifier of PCA dimensionality reduction and artificial neural network.
As optimization, the formation of each Haar-like feature is to be made of the matrix square frame with different sizes and position, and each matrix square frame includes image information.
As optimization, each matrix square frame is divided into the several function piece.
As optimization, image projection image to be detected is to functional block.
Each functional block is encoded to a left side, on, the right side, down at the coordinate of the vector of unit length in text.
As optimization, the rectangle for 20 * 20 is shown by two 10x20 frame tables:
feature2x1_1 2
0.00 0.00 0.50 1.00 1
0.50 0.00 1.00 1.00 -1 。
As optimization, the Haar feature that the rectangle for 20 * 20 is made up of four 10 * 10 rectangle is expressed as:
feature2x2_1 4
0.00 0.00 0.50 0.50 1
0.50 0.00 1.00 0.50 -1
0.00 0.50 0.50 1.00 -1
0.50 0.50 1.00 1.00 1 。
As optimization, in the division of integral image, division principle is a division methods of using maximum piece under the situation that guarantees picture quality.
As optimization, use the rectangular characteristic that is no more than 100 dimensions.
Compared with prior art, the invention has the beneficial effects as follows: by optimization process, accelerated data processing speed, realized fast detecting, and added choosing of separation vessel to improve adaptability for increasing accuracy and practicality to object to be measured to integral image.
Embodiment
In order to make purpose of the present invention, technical scheme and advantage clearer,, the present invention is further elaborated below in conjunction with embodiment.Should be appreciated that specific embodiment described herein only in order to explanation the present invention, and be not used in qualification the present invention.
Disclosed all features in this instructions except the feature of mutual eliminating, all can make up by any way.
Disclosed arbitrary feature in this instructions (comprising any accessory claim, summary) is unless special narration all can be replaced by other equivalences or the alternative features with similar purpose.That is, unless special narration, each feature is an example in a series of equivalences or the similar characteristics.
A kind of quick Haar wavelet character object detection method is characterized in that:
Step 1, use Haar-like feature are optimized integral image, realize the fast detecting to object;
Step 2, use are carried out dimensionality reduction based on the cascade classifier of PCA dimensionality reduction and artificial neural network.
The formation of each Haar-like feature is to be made of the matrix square frame with different sizes and position, and each matrix square frame includes image information.
Each matrix square frame is divided into the several function piece.
Image projection image to be detected is to functional block.
In the division of integral image, division principle is a division methods of using maximum piece under the situation that guarantees picture quality.
First step to Flame Image Process is an integral image, and integral image is used to extract quick Haar-like feature.Extract in the Pixel-level space in original image.
The formation of each Haar-like feature is to be made of the square frame rectangle with different sizes and position, and a square frame rectangle comprises some image informations.For example: consider some rectangles of 20 * 20, rectangle is that the size by two rectangles of the inside is that the size of 10 * 20 or four rectangles is that 10 * 10 piece constitutes.Here said is the base unit of image processing field, is other minimum treat unit of image pixel-class, and according to the clear size of carrying out definition block of reality, and these pieces are exactly functional block.The current function that to formulate such function collection complete more than be functional block disposes all possible combination successively.
Basis with rectangle function of 20 * 20 of this feature is that the projection testing image is to this functional block, and the function that this projection function constitutes is gathered, here with the above-mentioned integral image of mentioning very big relation is arranged, on the division of integral image, we illustrate: if the size of 2 10 * 20 such rectangles is equal to the size of 4 10 * 10 rectangles, below we do some configurations to such piecemeal that may exist, the foundation of configuration is: the rectangle sample that existence may more access times, the piece of being divided is big more, the number of times that uses is few more, accordingly, required time of image detection is just short more.We detect for trying to achieve fast, will use the division methods of bigger piece under the situation that guarantees picture quality, promptly comparatively widely used rectangle sample.Like this, in follow-up image detection process, just in the process of projection a very little time quantum only takes place and reach our requirement.The assurance of picture quality needs to obtain through experiment in conjunction with actual conditions, determines actual used size then.
Secondly, use is no more than the rectangular characteristic detection method of 100 dimensions, make the still less consuming time of calculating, the image detection feature is more flexible, write the mode of file by the data that will train, in the process of choosing, only need to determine clear and definite file, just can determine to use concrete dimension, promptly the dimension of characteristic draws in advance.
Obtain with the Haar characterization method that realizes different low dimensions and be:
Use continuous rectangular node to be divided into 1 * 2,2 * 1,2 * 2,3 * 1,1 * 3,3 * 2,2 * 3,4 * 5,5 * 4 square frames etc., they are encoded as a left side, on, the right side, down at the coordinate of the vector of unit length in text.Therefore, above-mentioned 20 * 20 the rectangle of mentioning is shown by two 10x20 frame tables:
feature2x1_1 2
0.00 0.00 0.50 1.00 1
0.50 0.00 1.00 1.00 -1
If the Haar feature that 4 10 * 10 rectangle is formed is expressed as:
feature2x2_1 4
0.00 0.00 0.50 0.50 1
0.50 0.00 1.00 0.50 -1
0.00 0.50 0.50 1.00 -1
0.50 0.50 1.00 1.00 1
In actual and follow-up processing procedure, the Haar feature is the basis that quantizes the PCA dimensionality reduction fast, and the good detection precision is arranged, and the Haar feature becomes the basic data structure of the present invention fast.
People's face test section based on quick Haar feature, in the object or people's face test section of reality, the Haar mark sheet has revealed higher detection process treatment characteristic fast, especially shows the speed aspect of detection, and processing speed can reach the speed of 75-100 frame/second.Be directed to the detection of people's face and non-face part, we can obtain higher detection speed.
In the computation process of using quick Haar feature, the summation method to the optimization rear space rectangle of this proposition gets a promotion the integral image disposal route on processing speed.
Use is no more than the rectangular characteristic of 100 dimensions.
For the excessive feature of quantity, make processing speed slack-off near flood tide rectangular characteristic as the astronomical figure, use one to be no more than 100 dimension rectangular characteristic and to send into PCA dimensionality reduction space as the eigenface model and realize final fast detecting effect here.
Use the cascade classifier of quick Haar feature
For the speed of accelerating to detect, use the cascade classifier of neural network.Also can train and add as SVM, existing sorter class libraries such as KNN, each sorter comprises a non-linear separate confinement at least, actual classifying quality is in order to indicate the classification performance in cross-validation process, use has the code of sorter of the neural network of optimization, can produce better nicety of grading, obtain classifying quality preferably on a small scale at one.In test, use 9 sorters to handle all functions at last, and best classifying quality is provided.For increasing accuracy and practicality, first sorter only uses 3 quick Haar feature descriptions, is incremented to 100 the highest dimensional features successively.If it produces a negative classification results, it is non-detected object that the result is considered to, and uses next sorter so.
Claims (9)
1. quick Haar wavelet character object detection method is characterized in that:
Step 1, use Haar-like feature are optimized integral image, realize the fast detecting to object;
Step 2, use are carried out dimensionality reduction based on the cascade classifier of PCA dimensionality reduction and artificial neural network.
2. a kind of quick Haar wavelet character object detection method according to claim 1 is characterized in that:
The formation of each Haar-like feature is to be made of the matrix square frame with different sizes and position, and each matrix square frame includes image information.
3. a kind of quick Haar wavelet character object detection method according to claim 2 is characterized in that:
Each matrix square frame is divided into the several function piece.
4. a kind of quick Haar wavelet character object detection method according to claim 3 is characterized in that:
Image projection image to be detected is to functional block.
5. a kind of quick Haar wavelet character object detection method according to claim 4 is characterized in that:
Each functional block is encoded to a left side, on, the right side, down at the coordinate of the vector of unit length in text.
6. a kind of quick Haar wavelet character object detection method according to claim 5 is characterized in that:
Rectangle for 20 * 20 is shown by two 10x20 frame tables:
feature2x1_1 2
0.00 0.00 0.50 1.00 1
0.50 0.00 1.00 1.00 -1 。
7. a kind of quick Haar wavelet character object detection method according to claim 5 is characterized in that:
The Haar feature that rectangle for 20 * 20 is made up of four 10 * 10 rectangle is expressed as:
feature2x2_1 4
0.00 0.00 0.50 0.50 1
0.50 0.00 1.00 0.50 -1
0.00 0.50 0.50 1.00 -1
0.50 0.50 1.00 1.00 1 。
8. a kind of quick Haar wavelet character object detection method according to claim 3 is characterized in that:
In the division of integral image, division principle is a division methods of using maximum piece under the situation that guarantees picture quality.
9. a kind of quick Haar wavelet character object detection method according to claim 1 is characterized in that:
Use is no more than the rectangular characteristic of 100 dimensions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013101047392A CN103226711A (en) | 2013-03-28 | 2013-03-28 | Quick Haar wavelet feature object detecting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013101047392A CN103226711A (en) | 2013-03-28 | 2013-03-28 | Quick Haar wavelet feature object detecting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103226711A true CN103226711A (en) | 2013-07-31 |
Family
ID=48837152
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2013101047392A Pending CN103226711A (en) | 2013-03-28 | 2013-03-28 | Quick Haar wavelet feature object detecting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103226711A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104166840A (en) * | 2014-07-22 | 2014-11-26 | 厦门亿联网络技术股份有限公司 | Focusing realization method based on video conference system |
| CN104615147A (en) * | 2015-02-13 | 2015-05-13 | 中国北方车辆研究所 | Method and system for accurately positioning polling target of transformer substation |
| RU2634374C2 (en) * | 2015-11-17 | 2017-10-26 | Закрытое акционерное общество "Радиотехнические и Информационные Системы воздушно-космической обороны" (ЗАО "РТИС ВКО") | Method of optical detecting low-contrast dynamic objects on complex atmospheric background |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110142345A1 (en) * | 2009-12-14 | 2011-06-16 | Electronics And Telecommunications Research Institute | Apparatus and method for recognizing image |
| CN102254151A (en) * | 2011-06-16 | 2011-11-23 | 清华大学 | Driver fatigue detection method based on face video analysis |
| CN102722736A (en) * | 2012-06-13 | 2012-10-10 | 合肥工业大学 | Method for splitting and identifying character strings at complex interference |
| US20120328160A1 (en) * | 2011-06-27 | 2012-12-27 | Office of Research Cooperation Foundation of Yeungnam University | Method for detecting and recognizing objects of an image using haar-like features |
| CN102855496A (en) * | 2012-08-24 | 2013-01-02 | 苏州大学 | Method and system for authenticating shielded face |
-
2013
- 2013-03-28 CN CN2013101047392A patent/CN103226711A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110142345A1 (en) * | 2009-12-14 | 2011-06-16 | Electronics And Telecommunications Research Institute | Apparatus and method for recognizing image |
| CN102254151A (en) * | 2011-06-16 | 2011-11-23 | 清华大学 | Driver fatigue detection method based on face video analysis |
| US20120328160A1 (en) * | 2011-06-27 | 2012-12-27 | Office of Research Cooperation Foundation of Yeungnam University | Method for detecting and recognizing objects of an image using haar-like features |
| CN102722736A (en) * | 2012-06-13 | 2012-10-10 | 合肥工业大学 | Method for splitting and identifying character strings at complex interference |
| CN102855496A (en) * | 2012-08-24 | 2013-01-02 | 苏州大学 | Method and system for authenticating shielded face |
Non-Patent Citations (1)
| Title |
|---|
| 穆志纯等: "《生物特征识别技术 人耳自动识别》", 31 December 2012 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104166840A (en) * | 2014-07-22 | 2014-11-26 | 厦门亿联网络技术股份有限公司 | Focusing realization method based on video conference system |
| CN104615147A (en) * | 2015-02-13 | 2015-05-13 | 中国北方车辆研究所 | Method and system for accurately positioning polling target of transformer substation |
| RU2634374C2 (en) * | 2015-11-17 | 2017-10-26 | Закрытое акционерное общество "Радиотехнические и Информационные Системы воздушно-космической обороны" (ЗАО "РТИС ВКО") | Method of optical detecting low-contrast dynamic objects on complex atmospheric background |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | WLBP: Weber local binary pattern for local image description | |
| Ren et al. | Object detection using edge histogram of oriented gradient | |
| Moghaddasi et al. | SVD-based image splicing detection | |
| Pai et al. | A novel crowd density estimation technique using local binary pattern and Gabor features | |
| Mena et al. | Comparative study of the features used by algorithms based on viola and jones face detection algorithm | |
| Gonzalez-Reyna et al. | Eigen‐Gradients for Traffic Sign Recognition | |
| CN103226711A (en) | Quick Haar wavelet feature object detecting method | |
| Cotter et al. | A hardware accelerated multilevel visual classifier for embedded visual-assist systems | |
| Liu et al. | Low resolution pedestrian detection using light robust features and hierarchical system | |
| Reddy et al. | Face recognition using multi region prominent lbp representation | |
| Zagoris et al. | Scene text detection on images using cellular automata | |
| Thanh Do et al. | Facial feature extraction using geometric feature and independent component analysis | |
| Zhao et al. | Facial expression recognition based on local binary patterns and least squares support vector machines | |
| Megahed et al. | Face2face manipulation detection based on histogram of oriented gradients | |
| Kim et al. | An improved license plate recognition technique in outdoor image | |
| Shokoohi et al. | Expression recognition using directional gradient local pattern and gradient-based ternary texture patterns | |
| Terauchi et al. | Analysis Based on Distributed Representations of Various Parts Images in Four-Scene Comics Story Dataset | |
| Wang et al. | Face recognition algorithm based on Haar-like features and gentle AdaBoost feature selection via sparse representation | |
| Crudge et al. | Landmark recognition using machine learning | |
| Wu et al. | Face Recognition based on LBP and LNMF Algorithm | |
| Das et al. | Design of pedestrian detectors using combinations of scale spaces and classifiers | |
| Hui-bin et al. | Recognition of individual object in focus people group based on deep learning | |
| Ensari et al. | Occluded face recognition using correntropy-based nonnegative matrix factorization | |
| Moghaddasi et al. | Image splicing detection using singular value decomposition | |
| Bazmi et al. | Increasing the Accuracy of Detection and Recognition in Visual Surveillance. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130731 |