| Collection Title: | SIU Thesis | | Title : | The Application Research of Cellular Neural Networks (CNN) in Image Processing | | Material Type: | printed text | | Authors: | Gangyi Hu, Author ; Sumeth Yuenyong, Associated Name ; Myint, Lin Min Min, Associated Name | | Publisher: | Bangkok: Shinawatra University | | Publication Date: | 2017 | | Pagination: | x, 172 p. | | Layout: | ill, Tables | | Size: | 30 cm. | | Price: | 500.00 | | General note: | SIU THE: SOST-PhD-IT-2017-01
Thesis. [PhD [Information Technology]]. -- Shinawatra University, 2017 | | Languages : | English (eng) | | Descriptors: | [LCSH]Genetic algorithms
[LCSH]Image processing
| | Keywords: | Cellular Neural Networks,
Image Processing,
Edge Detection,
Genetic Algorithm,
Denoising,
Target Protection,
Chaos,
Encryption,
Asymmetric Algorithm | | Abstract: | The cellular neural networks (CNN) are composed of many cell units which are the local connection. Each cell is consisting of linear and nonlinear circuits. This structure can be realized as very large-scale integrated circuit (VLSI), which can be used in large-scale parallel computing. Therefore, the cellular neural networks can be applied to solve the problem such as image processing, signal processing, robot and biological vision. It is one of the hot spots in the field of neural networks.
This thesis mainly studies the practical problems of image processing application, and it uses the advantages of cellular neural networks, such as nonlinear and high-speed, real-time parallel computing. And then it is combined with the genetic training algorithm, filtering analysis, chaos theory and modern cryptography theory to solve some problems in image processing, which expands the application scope of cellular neural networks.
The main research works of this thesis are as below.
1) It reviews the basic concepts, background, development status, and hardware realization of the cellular neural networks, and then it analyzes the dynamic range and stability characteristics of the cellular neural networks and expounds the principle and significance of cellular neural networks for image processing.
2) In the research on edge detection for the infrared image. It studies the principle, purpose, and significance of infrared image edge detection, and then compared with the traditional target edge detection algorithm such as the Canny algorithm for the infrared image. At last, it discusses using the genetic algorithm with cellular neural networks for infrared image edge detection. It proposes an algorithm to get the cellular neural networks template parameters; this algorithm is based on subpopulation genetic algorithm. Through the optimum design of filial populations as well as the improvement of the parallel genetic offspring, it overcomes the disadvantages such as premature
convergence of simple genetic algorithm to create template parameters, and the fitness function is developed by using the Lyapunov function. The experiments show that this improved genetic algorithm combined with cellular neural networks is used to detect the edge of the infrared image, which can get the edge of infrared image accurately and quickly. This algorithm has fast convergence speed and accurate target edge detection.
3) The image denoising is an important research aspect in image processing, aiming at the contradiction between denoising and edge preserving information in the traditional denoising method. It first studies the noise model, the principle and the evaluation criteria of image denoising, then according to the characteristics of the three templates of cellular neural networks; it proposes an edge constraint adaptive filtering algorithm based on cellular neural networks for image denoising. In the process of designing the three templates separately in cellular neural networks, the control template references the advantage of spatial filtering denoising, it resembles spatial domain denoising filter. The feedback template sets as a matrix which generated by a high pass filter to achieve edge preservation. Thus, it can not only perform denoising but also can protect edges in an image. In the process of designing the threshold template, it uses the different gray levels in an image to achieve the threshold adjustment adaptively. The experiments show that this algorithm has best denoising effect for various image noise types. When comparing the edge protection effect with other denoising algorithms, it can protect the edge information very well. The Peak signal to noise ratio (PSNR) is also higher than other traditional denoising algorithms. This algorithm is mainly developing a new method for image denoising.
4) The image encryption is a vital part of image transmission and an important guarantee to prevent the leakage of image information. It introduces the principle of image encryption and the method of generating hyper chaotic system based on cellular neural networks, then using these hyper chaotic sequences from the chaotic system for image encryption. The image encryption mainly includes two steps, one is changing the image pixel position, and the other is replacing the pixel values. This algorithm is based on cellular neural networks six dimensional hyper chaotic systems. The main idea of
the algorithm is that the image encryption key is first to been input to the cellular neural networks system to generate six-dimensional chaotic sequences, and then encrypt the key by an asymmetric encryption algorithm. In the process of encryption image, firstly change the original image pixel positions according to the chaotic sequences, and then put the pixel shuffled image and the modified chaotic sequences through the XOR operation to replace the pixel value. Thus, it can get the final cipher image. The image decryption process is the reverse of the above two steps. This algorithm has two advantages. First, it realizes the encryption and protection of the image based on the hyper chaotic sequences generated by the high dimensional of cellular neural networks. Secondly, the key is encrypted by asymmetric encryption algorithm such as RSA algorithm. It can protect the key in the transmission process. The experimental results show that this algorithm can achieve low correlation between pixels, and its cipher image can also achieve high change rate of pixel ratio, high information entropy, and strong anti-hacking ability. Compared with other chaotic image encryption schemes, this algorithm has substantial practical value. | | Curricular : | BSCS/MSIT/PhDT | | Record link: | http://libsearch.siu.ac.th/siu/opac_css/index.php?lvl=notice_display&id=27204 |
SIU Thesis. The Application Research of Cellular Neural Networks (CNN) in Image Processing [printed text] / Gangyi Hu, Author ; Sumeth Yuenyong, Associated Name ; Myint, Lin Min Min, Associated Name . - [S.l.] : Bangkok: Shinawatra University, 2017 . - x, 172 p. : ill, Tables ; 30 cm. 500.00 SIU THE: SOST-PhD-IT-2017-01
Thesis. [PhD [Information Technology]]. -- Shinawatra University, 2017 Languages : English ( eng) | Descriptors: | [LCSH]Genetic algorithms
[LCSH]Image processing
| | Keywords: | Cellular Neural Networks,
Image Processing,
Edge Detection,
Genetic Algorithm,
Denoising,
Target Protection,
Chaos,
Encryption,
Asymmetric Algorithm | | Abstract: | The cellular neural networks (CNN) are composed of many cell units which are the local connection. Each cell is consisting of linear and nonlinear circuits. This structure can be realized as very large-scale integrated circuit (VLSI), which can be used in large-scale parallel computing. Therefore, the cellular neural networks can be applied to solve the problem such as image processing, signal processing, robot and biological vision. It is one of the hot spots in the field of neural networks.
This thesis mainly studies the practical problems of image processing application, and it uses the advantages of cellular neural networks, such as nonlinear and high-speed, real-time parallel computing. And then it is combined with the genetic training algorithm, filtering analysis, chaos theory and modern cryptography theory to solve some problems in image processing, which expands the application scope of cellular neural networks.
The main research works of this thesis are as below.
1) It reviews the basic concepts, background, development status, and hardware realization of the cellular neural networks, and then it analyzes the dynamic range and stability characteristics of the cellular neural networks and expounds the principle and significance of cellular neural networks for image processing.
2) In the research on edge detection for the infrared image. It studies the principle, purpose, and significance of infrared image edge detection, and then compared with the traditional target edge detection algorithm such as the Canny algorithm for the infrared image. At last, it discusses using the genetic algorithm with cellular neural networks for infrared image edge detection. It proposes an algorithm to get the cellular neural networks template parameters; this algorithm is based on subpopulation genetic algorithm. Through the optimum design of filial populations as well as the improvement of the parallel genetic offspring, it overcomes the disadvantages such as premature
convergence of simple genetic algorithm to create template parameters, and the fitness function is developed by using the Lyapunov function. The experiments show that this improved genetic algorithm combined with cellular neural networks is used to detect the edge of the infrared image, which can get the edge of infrared image accurately and quickly. This algorithm has fast convergence speed and accurate target edge detection.
3) The image denoising is an important research aspect in image processing, aiming at the contradiction between denoising and edge preserving information in the traditional denoising method. It first studies the noise model, the principle and the evaluation criteria of image denoising, then according to the characteristics of the three templates of cellular neural networks; it proposes an edge constraint adaptive filtering algorithm based on cellular neural networks for image denoising. In the process of designing the three templates separately in cellular neural networks, the control template references the advantage of spatial filtering denoising, it resembles spatial domain denoising filter. The feedback template sets as a matrix which generated by a high pass filter to achieve edge preservation. Thus, it can not only perform denoising but also can protect edges in an image. In the process of designing the threshold template, it uses the different gray levels in an image to achieve the threshold adjustment adaptively. The experiments show that this algorithm has best denoising effect for various image noise types. When comparing the edge protection effect with other denoising algorithms, it can protect the edge information very well. The Peak signal to noise ratio (PSNR) is also higher than other traditional denoising algorithms. This algorithm is mainly developing a new method for image denoising.
4) The image encryption is a vital part of image transmission and an important guarantee to prevent the leakage of image information. It introduces the principle of image encryption and the method of generating hyper chaotic system based on cellular neural networks, then using these hyper chaotic sequences from the chaotic system for image encryption. The image encryption mainly includes two steps, one is changing the image pixel position, and the other is replacing the pixel values. This algorithm is based on cellular neural networks six dimensional hyper chaotic systems. The main idea of
the algorithm is that the image encryption key is first to been input to the cellular neural networks system to generate six-dimensional chaotic sequences, and then encrypt the key by an asymmetric encryption algorithm. In the process of encryption image, firstly change the original image pixel positions according to the chaotic sequences, and then put the pixel shuffled image and the modified chaotic sequences through the XOR operation to replace the pixel value. Thus, it can get the final cipher image. The image decryption process is the reverse of the above two steps. This algorithm has two advantages. First, it realizes the encryption and protection of the image based on the hyper chaotic sequences generated by the high dimensional of cellular neural networks. Secondly, the key is encrypted by asymmetric encryption algorithm such as RSA algorithm. It can protect the key in the transmission process. The experimental results show that this algorithm can achieve low correlation between pixels, and its cipher image can also achieve high change rate of pixel ratio, high information entropy, and strong anti-hacking ability. Compared with other chaotic image encryption schemes, this algorithm has substantial practical value. | | Curricular : | BSCS/MSIT/PhDT | | Record link: | http://libsearch.siu.ac.th/siu/opac_css/index.php?lvl=notice_display&id=27204 |
|
Add the result to your basket Make a suggestion Refine your search Apply to external sourcesSIU RS. Evaluation of Object Tracking Methods in OpenCV / Ithipol Kaosucharit / Bangkok: Shinawatra University - 2017
