CN104112113A - Improved characteristic convolutional neural network image identification method - Google Patents
Improved characteristic convolutional neural network image identification method Download PDFInfo
- Publication number
- CN104112113A CN104112113A CN201310137496.2A CN201310137496A CN104112113A CN 104112113 A CN104112113 A CN 104112113A CN 201310137496 A CN201310137496 A CN 201310137496A CN 104112113 A CN104112113 A CN 104112113A
- Authority
- CN
- China
- Prior art keywords
- image
- convolutional neural
- neural networks
- feature
- neural network
- 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)
- Image Processing (AREA)
Abstract
The invention discloses an improved characteristic convolutional neural network image identification method. According to the improved characteristic convolutional neural network image identification method, firstly, pre-processing on an image to be inputted is carried out; secondly, a characteristic extraction layer is added to a convolutional neural network structure, and the characteristics are amplified; image data with characteristic amplification is inputted to a convolutional neural network, convolutional operation learning identification for the characteristics in the image is carried out, output of the convolutional neural network is acquired, and offset operation on the image after pre-processing in the first step is further carried out to acquire textural characteristics of the image; lastly, the textural characteristics are analyzed to output a result, and outputs of the convolutional neural network are fused to output a final result. By adding the characteristic extraction layer, the characteristics of the image are reinforced, so extraction of some unconspicuous characteristics is facilitated, the image identification rate can be effectively improved, and methods can be further provided for image searching and image identification, and identification efficiency is improved.
Description
Technical field
The invention belongs to mode identification technology, particularly a kind of based on modified feature convolutional neural networks image-recognizing method.
Background technology
At present, pattern-recognition is still one of the most popular research direction of computer science.Giving computing machine sensory faculty, is the study hotspot of pattern-recognition.Scholars have proposed many algorithms for graph and image processing, the directions such as speech processes.
Aspect image recognition, be mainly from image, to extract unique point, then utilize this feature to identify image.At present, the main algorithm that is used to image recognition is a lot, such as: wavelet analysis, support vector machine (SVM), genetic algorithm, neural network algorithm etc.
Heisele, Bernd has proposed a kind of face recognition algorithm based on SVM.It is divided into respectively a plurality of regions by people's face, utilize face image data storehouse to train algorithm, then from these regions, extract different features for image recognition, obtained certain effect (Heisele, Bernd. Face recognition with support vector machines:global versus component-based approach[C]. IEEE, 2001:688-694.).
Aspect neural network, in the sixties in 20th century, Hubel and Wiesel find that its unique network structure can reduce the complicacy of Feedback Neural Network effectively during for the neuron of local sensitivity and set direction in research cat cortex, have then proposed convolutional neural networks (Convolutional Neural Networks-is called for short CNN).Along with the deep understanding of scholar for convolutional neural networks, according to the difference of application scenarios, scholars have proposed some weakness that different modified versions overcame convolutional neural networks originally.
Aspect convolutional neural networks, Yann LeCun etc. has proposed to utilize convolutional neural networks to carry out the identification of hand-written script.But because the complicacy of font is higher, so the Algorithm for Training time is longer, obtained certain effect (Y.LeCun, L.Bottou, Y.Bengio, andP.Haffner. " Gradient-based Learning Applied to Document Recognition ", Proceedings of the IEEE, vol.86, pp.2278 – 2324, November1998.).
Summary of the invention
1, object of the present invention.
The invention provides and a kind ofly based on modified feature convolutional neural networks image-recognizing method, can in convolutional neural networks, for feature identifying, instruct, strengthen its ability in feature extraction, improve image recognition rate.
2, the technical solution adopted in the present invention.
Based on modified feature convolutional neural networks image-recognizing method, adopt following steps:
The first step, carries out pre-service to image to be entered, gets rid of the image of serious non-compliance regulation, and according to the attribute of input picture, design convolutional neural networks structure;
Second step adds feature extraction layer to strengthen convolutional neural networks for the extractability of characteristics of image in convolutional neural networks structure, and feature is amplified;
The 3rd step, the feature enlarged image data input convolutional neural networks by described step 2, carries out convolution operation study identification to the feature in image, obtains the output of convolutional neural networks;
The 4th step, carries out offset operation by pretreated image in described first step, obtains the textural characteristics of image;
The 5th step, to Output rusults after the analysis of texture of described step 4, and merges the output of the convolutional neural networks in described step 3, obtains net result and exports.
3, beneficial effect of the present invention.
The present invention compared with prior art, its remarkable advantage: 1), by increasing feature extraction layer, can instruct for the characteristic extraction procedure of convolutional neural networks, strengthened ability in feature extraction; 2) image for input does not need too much artificial pre-service.
Below in conjunction with accompanying drawing, the present invention is described in further detail.
Accompanying drawing explanation
Fig. 1 is feature convolutional neural networks structural drawing.
Fig. 2 is the skew schematic diagram of texture feature extraction pixel.
Fig. 3 is texture feature extraction weight matrix schematic diagram.
Embodiment
Embodiment
As shown in Figure 1, the step of employing is as follows for general structure of the present invention:
The first step, carries out pre-service to view data, manually rejects the vicious image of obvious tool, according to the attribute design convolutional neural networks structure of image, and realizes its algorithm.
Second step, according to pretreated view data, analyzes its characteristic properties, and its feature is amplified.In the present embodiment, unique point is not special obviously by the method for transform color space, its feature to be amplified in rgb space, by original RGB data transformation to HLS color space.
Transformation for mula is as follows:
The 3rd step, the view data input convolutional neural networks by extension, carries out convolution operation to it, and finally obtains network output N
1.
The 4th step, by former figure
to 8 directions, be respectively offset a pixel, as shown in Figure 2, produce respectively new figure
?
,
with the new figure producing
,
right
operate and obtain new figure
, computing formula is as follows:,
For can basis
,
this 8 secondary figure extrapolates the position relationship between itself and former figure, adopts 3 * 3 matrixes
calculate.
as shown in Figure 3, by funtcional relationship f to 8 each pixel of secondary picture
operate, computing formula is as follows:
If 32 * 32 of general pattern sizes are number of features totally 1024 features now, number of features is too much, the inconvenient analysis to feature, and in the new feature figure now obtaining, element size is all between 1～255, so now gained characteristic pattern is at 1～255 enterprising column hisgram
, number of features reduces to 255, is convenient to obtain Local textural feature.
The 5th step, need to analyze 255 Local textural features that extract here, and Output rusults is N
2, finally to N
2merge the N of output with above-mentioned convolutional neural networks
1merge, thereby obtain last result.
Above-described embodiment does not limit the present invention in any way, and every employing is equal to replaces or technical scheme that the mode of equivalent transformation obtains all drops in protection scope of the present invention.
Claims (4)
1. based on a modified feature convolutional neural networks image-recognizing method, it is characterized in that adopting following steps:
The first step, carries out pre-service to image to be entered, gets rid of the image of serious non-compliance regulation, and according to the attribute of input picture, design convolutional neural networks structure;
Second step adds feature extraction layer to strengthen convolutional neural networks for the extractability of characteristics of image in convolutional neural networks structure, and feature is amplified;
The 3rd step, the feature enlarged image data input convolutional neural networks by described step 2, carries out convolution operation study identification to the feature in image, obtains the output of convolutional neural networks;
The 4th step, carries out offset operation by pretreated image in described first step, obtains the textural characteristics of image;
The 5th step, to Output rusults after the analysis of texture of described step 4, and merges the output of the convolutional neural networks in described step 3, obtains net result and exports.
2. modified feature convolutional neural networks according to claim 1, is characterized in that the feature amplification employing of described step 2 transforms to HLS color space by unique point at rgb space, and transform method is as follows:
。
3. modified feature convolutional neural networks according to claim 1, it is characterized in that the 4th described step image being carried out to offset operation, to obtain the textural characteristics method of image as follows:
Step 3.1, by former figure
to 8 directions, be respectively offset a pixel, produce respectively new figure
?
,
, with the new figure producing
?
,
right
operate and obtain new figure
, method is as follows:
Step 3.2,
?
,
with former figure
position relationship adopt 3 * 3 matrixes
calculate matrix
as shown in form,
By funtcional relationship f to 8 each pixel of secondary picture
operate, computing formula is as follows:
。
4. modified feature convolutional neural networks according to claim 3, in the new feature figure that it is characterized in that obtaining, element size is all between 1～255, to gained characteristic pattern, at 1～255 enterprising column hisgram, obtaining number of features is 255.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310137496.2A CN104112113A (en) | 2013-04-19 | 2013-04-19 | Improved characteristic convolutional neural network image identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310137496.2A CN104112113A (en) | 2013-04-19 | 2013-04-19 | Improved characteristic convolutional neural network image identification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104112113A true CN104112113A (en) | 2014-10-22 |
Family
ID=51708898
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310137496.2A Pending CN104112113A (en) | 2013-04-19 | 2013-04-19 | Improved characteristic convolutional neural network image identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104112113A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104517122A (en) * | 2014-12-12 | 2015-04-15 | 浙江大学 | Image target recognition method based on optimized convolution architecture |
| CN104679863A (en) * | 2015-02-28 | 2015-06-03 | 武汉烽火众智数字技术有限责任公司 | Method and system for searching images by images based on deep learning |
| CN105389596A (en) * | 2015-12-21 | 2016-03-09 | 长沙网动网络科技有限公司 | Method for enabling convolutional neural network to be suitable for recognition of pictures of various sizes |
| CN105760442A (en) * | 2016-02-01 | 2016-07-13 | 中国科学技术大学 | Image feature enhancing method based on database neighborhood relation |
| CN106408086A (en) * | 2016-09-12 | 2017-02-15 | 上海影城有限公司 | Deep learning neural network processing method and deep learning neural network processing system for image optimization |
| CN106650781A (en) * | 2016-10-21 | 2017-05-10 | 广东工业大学 | Convolutional neural network image recognition method and device |
| CN107292319A (en) * | 2017-08-04 | 2017-10-24 | 广东工业大学 | The method and device that a kind of characteristic image based on deformable convolutional layer is extracted |
| CN107563390A (en) * | 2017-08-29 | 2018-01-09 | 苏州智萃电子科技有限公司 | A kind of image-recognizing method and system |
| CN107688784A (en) * | 2017-08-23 | 2018-02-13 | 福建六壬网安股份有限公司 | A kind of character identifying method and storage medium based on further feature and shallow-layer Fusion Features |
| CN107729948A (en) * | 2017-10-31 | 2018-02-23 | 京东方科技集团股份有限公司 | Image processing method and device, computer product and storage medium |
| CN107784678A (en) * | 2017-11-08 | 2018-03-09 | 北京奇虎科技有限公司 | Generation method, device and the terminal of cartoon human face image |
| CN108446724A (en) * | 2018-03-12 | 2018-08-24 | 江苏中天科技软件技术有限公司 | A kind of fusion feature sorting technique |
| CN108830322A (en) * | 2018-06-15 | 2018-11-16 | 联想(北京)有限公司 | A kind of image processing method and device, equipment, storage medium |
| CN110751162A (en) * | 2018-07-24 | 2020-02-04 | 杭州海康威视数字技术股份有限公司 | Image identification method and device and computer equipment |
| CN111598856A (en) * | 2020-05-08 | 2020-08-28 | 浙江工商大学 | Chip surface defect automatic detection method and system based on defect-oriented multi-point positioning neural network |
| US10810491B1 (en) | 2016-03-18 | 2020-10-20 | Amazon Technologies, Inc. | Real-time visualization of machine learning models |
| US11875250B1 (en) | 2017-06-19 | 2024-01-16 | Amazon Technologies, Inc. | Deep neural networks with semantically weighted loss functions |
-
2013
- 2013-04-19 CN CN201310137496.2A patent/CN104112113A/en active Pending
Non-Patent Citations (1)
| Title |
|---|
| 陆璐: "卷积神经网络的研究及其在车牌识别系统中的应用", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104517122A (en) * | 2014-12-12 | 2015-04-15 | 浙江大学 | Image target recognition method based on optimized convolution architecture |
| CN104679863B (en) * | 2015-02-28 | 2018-05-04 | 武汉烽火众智数字技术有限责任公司 | It is a kind of based on deep learning to scheme to search drawing method and system |
| CN104679863A (en) * | 2015-02-28 | 2015-06-03 | 武汉烽火众智数字技术有限责任公司 | Method and system for searching images by images based on deep learning |
| CN105389596A (en) * | 2015-12-21 | 2016-03-09 | 长沙网动网络科技有限公司 | Method for enabling convolutional neural network to be suitable for recognition of pictures of various sizes |
| CN105389596B (en) * | 2015-12-21 | 2018-05-29 | 长沙网动网络科技有限公司 | The method that convolutional neural networks are suitable for identification sizes picture |
| CN105760442A (en) * | 2016-02-01 | 2016-07-13 | 中国科学技术大学 | Image feature enhancing method based on database neighborhood relation |
| CN105760442B (en) * | 2016-02-01 | 2019-04-26 | 中国科学技术大学 | Characteristics of image Enhancement Method based on database neighborhood relationships |
| US10810491B1 (en) | 2016-03-18 | 2020-10-20 | Amazon Technologies, Inc. | Real-time visualization of machine learning models |
| CN106408086A (en) * | 2016-09-12 | 2017-02-15 | 上海影城有限公司 | Deep learning neural network processing method and deep learning neural network processing system for image optimization |
| CN106650781A (en) * | 2016-10-21 | 2017-05-10 | 广东工业大学 | Convolutional neural network image recognition method and device |
| US11875250B1 (en) | 2017-06-19 | 2024-01-16 | Amazon Technologies, Inc. | Deep neural networks with semantically weighted loss functions |
| CN107292319A (en) * | 2017-08-04 | 2017-10-24 | 广东工业大学 | The method and device that a kind of characteristic image based on deformable convolutional layer is extracted |
| CN107688784A (en) * | 2017-08-23 | 2018-02-13 | 福建六壬网安股份有限公司 | A kind of character identifying method and storage medium based on further feature and shallow-layer Fusion Features |
| CN107563390A (en) * | 2017-08-29 | 2018-01-09 | 苏州智萃电子科技有限公司 | A kind of image-recognizing method and system |
| CN107729948A (en) * | 2017-10-31 | 2018-02-23 | 京东方科技集团股份有限公司 | Image processing method and device, computer product and storage medium |
| US12094193B2 (en) | 2017-10-31 | 2024-09-17 | Boe Technology Group Co., Ltd. | Image processing method and image processing device |
| US11138738B2 (en) | 2017-10-31 | 2021-10-05 | Boe Technology Group Co., Ltd. | Image processing method and image processing device |
| CN107784678B (en) * | 2017-11-08 | 2021-11-12 | 北京奇虎科技有限公司 | Cartoon face image generation method and device and terminal |
| CN107784678A (en) * | 2017-11-08 | 2018-03-09 | 北京奇虎科技有限公司 | Generation method, device and the terminal of cartoon human face image |
| CN108446724A (en) * | 2018-03-12 | 2018-08-24 | 江苏中天科技软件技术有限公司 | A kind of fusion feature sorting technique |
| CN108830322A (en) * | 2018-06-15 | 2018-11-16 | 联想(北京)有限公司 | A kind of image processing method and device, equipment, storage medium |
| CN110751162A (en) * | 2018-07-24 | 2020-02-04 | 杭州海康威视数字技术股份有限公司 | Image identification method and device and computer equipment |
| CN111598856A (en) * | 2020-05-08 | 2020-08-28 | 浙江工商大学 | Chip surface defect automatic detection method and system based on defect-oriented multi-point positioning neural network |
| CN111598856B (en) * | 2020-05-08 | 2024-05-10 | 无锡瑞特思普智能科技有限公司 | Chip surface defect automatic detection method and system based on defect-oriented multipoint positioning neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104112113A (en) | Improved characteristic convolutional neural network image identification method | |
| Aptoula et al. | Deep learning with attribute profiles for hyperspectral image classification | |
| CN107273800B (en) | Attention mechanism-based motion recognition method for convolutional recurrent neural network | |
| Reyes et al. | Fine-tuning Deep Convolutional Networks for Plant Recognition. | |
| Khairnar et al. | Disease detection and diagnosis on plant using image processing—a review | |
| CN110321967B (en) | Image classification improvement method based on convolutional neural network | |
| CN108764242A (en) | Off-line Chinese Character discrimination body recognition methods based on deep layer convolutional neural networks | |
| CN111178312B (en) | Face expression recognition method based on multi-task feature learning network | |
| CN107330387B (en) | Pedestrian detection method based on image data | |
| CN104794455A (en) | Dongba hieroglyphic recognizing method | |
| CN115018039A (en) | Neural network distillation method, target detection method and device | |
| Tsai et al. | A single‐stage face detection and face recognition deep neural network based on feature pyramid and triplet loss | |
| KR100956747B1 (en) | Computer Architecture Combining Neural Network and Parallel Processor, and Processing Method Using It | |
| CN113297964B (en) | Video target recognition model and method based on deep migration learning | |
| Li et al. | FVGNN: A novel GNN to finger vein recognition from limited training data | |
| CN110598727B (en) | Model construction method based on transfer learning, image recognition method and device thereof | |
| CN103440332A (en) | Image searching method based on relation matrix regularization enhancement representation | |
| Yan et al. | Design and experiment of facial expression recognition method based on LBP and CNN | |
| Mehmeti-Göpel et al. | Nonlinear advantage: trained networks might not be as complex as you think | |
| Li et al. | Handwritten digit recognition via active belief decision trees | |
| Li et al. | Convolutional neural network based multi-focus image fusion | |
| CN106780637A (en) | A kind of fast parallel image thinning algorithm based on pulse nerve membranous system | |
| CN109635709B (en) | Facial expression recognition method based on significant expression change area assisted learning | |
| Kumar et al. | Red green blue depth image classification using pre-trained deep convolutional neural network | |
| Zong et al. | Application of convolutional neural network in target detection of millimeter wave imaging |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20141022 |
|
| WD01 | Invention patent application deemed withdrawn after publication |