The following is just my estimate:
The first topic should be related to circuit design. If you are good at circuit science and like debugging practical things, then this is for you.
The second one is to do experiments with software such as matlab. It is easy to get some experimental data and charts, and relevant papers are also easy to find.
On the whole, the second one will be simpler.
The following is bad information about the algorithm research of image blind restoration, but you can refer to it:
Topic: Research on blind image restoration algorithm.
Author: Zhan Zujian Date: February 27, 2009
Authors: Class 05 Communication Project of Mingde College of Guizhou University, Guiyang, Guizhou, 550003.
Title:
Science and Education Innovation of Kehai Story Expo
Abstract: When the point spread function is unknown or unknown, the process of restoring the original image from the observed degraded image is called blind image restoration. In recent years, blind image restoration algorithms have been widely studied. This paper introduces the present situation of blind image restoration algorithm, and further studies its development direction.
I. Introduction
Image restoration is a big field in image processing, which has a wide range of applications and is a research hotspot at present. The main purpose of image restoration is to make the degraded image go through certain processing, remove the degradation factors, and restore it to the original image with maximum fidelity. Traditional image restoration assumes that the image degradation model is known. However, in many cases, the image degradation model is unknown or has little prior knowledge, so the so-called blind restoration must be carried out. Because of its importance and arduousness, it has become a research hotspot. At present, the available observation images are the results of the real images imaged by the observation system. Due to the limitation of the physical characteristics of the observation system and the influence of the observation environment, there are inevitably deviations and distortions between the observed image and the real image, which is called that the observation system degrades the real image. The purpose of image restoration is to get the real image by analyzing and calculating the degraded observation image.
Second, the status quo of image blind restoration algorithms
Generally speaking, blind image restoration methods are mainly divided into the following two categories: first, the point spread function is estimated by using the special characteristics of the real image, and then with the help of the estimated point spread function, the image is restored by using the classical image restoration methods. This method divides the process of PSF estimation and image restoration into two different processes, so it has the characteristics of small calculation. Secondly, the point spread function identification and real image estimation are combined to identify the point spread function and real image at the same time. This algorithm is complex and requires a lot of calculation. In addition, the point spread function also considers the complex situation of spatial change. In view of the present situation of blind restoration algorithms, according to the characteristics of the degradation model, the algorithms are re-divided into three categories: space-invariant single-channel blind restoration algorithm, space-invariant multi-channel blind restoration algorithm and space-varying image blind restoration algorithm.
Blind restoration algorithm for (1) single-channel spatially invariant images.
Among these algorithms, parametric method and iterative method are the most commonly used.
1) parameter method. The so-called parameter method, that is, model parameter method, is to describe PSF and real image with a certain model, but the parameters of the model need to be identified. Among the parameter methods, there are two typical methods: prior fuzzy identification method and ARMA parameter estimation method. The former is a 1 type image blind restoration algorithm, so the computation is small. The latter identifies both PSF and real image model parameters, which belongs to the second kind of image blind restoration algorithm.
2) Iterative method. The so-called iterative method is a method to identify PSF and real image simultaneously through the iterative process of the algorithm and the constraints on the real image and PSF. The iterative method is single channel.
Image blind restoration algorithm is the most widely used algorithm. It does not need to establish a model and does not require PSF to be the minimum phase system, so it is closer to reality. Among these algorithms, iterative blind restoration algorithm (IBD) and recursive inverse filtering algorithm based on nonnegativity and decision domain (NAR2IF) are based on the minimization of higher-order statistical features.
Entropy algorithm is the most typical.
(2) Blind restoration of multi-channel 2D images
Multi-channel blind restoration of two-dimensional images extends the one-dimensional multi-channel blind source separation algorithm applied in digital communication field to two-dimensional situations for blind restoration of images. There are two algebraic methods for this algorithm. One is to identify the fuzzy function first, and then restore it with the conventional restoration algorithm. The other is the direct estimation inverse filter. The advantage of this algorithm is that it does not need to estimate the initial image, and there is no stability and convergence problem. The constraints on images and fuzzy functions are relaxed, and the algorithm is universal. However, the 1 algorithm requires the convergence of the algorithm to be restored; The second algorithm is sensitive to noise.
(3) Blind image restoration method with spatial variation.
In many practical applications, fuzziness often changes in space, but due to the difficulty of processing, there are few studies at present, and there are basically two kinds: correlation conversion recovery and direct method.
The basic idea of correlation transform restoration is region segmentation, that is, the whole image is divided into several local regions, and then the blur is assumed to be spatially invariant in each local region, and the image restoration algorithm with spatial invariance is used to restore it. These methods are all based on window fuzzy recognition technology, and image estimation depends on the size of the window. Because the fuzzy parameters are continuously changing, the assumption of spatial invariance in a large range is not established, so the accuracy of fuzzy estimation is poor, and this method can only deal with some spatially changing fuzzy images, which lacks universality. Secondly, there is a bell at the edge of the area.
The basic idea of direct method is to process the image directly. If the simplified two-dimensional recursive Kalman filter is used to directly convert the image model and the fuzzy model, its disadvantage is that it can only be used in limited models, and the calculation amount will increase significantly with the increase of the number of models; * * * yoke gradient iterative algorithm, but only a text image processing result of 3 1×3 1 is reported, and the effect of large image processing needs further study; Spatial change image system is established as Markov random model, and simulated annealing algorithm is used for maximum posterior estimation of restoration process. This method avoids windowing the image and can overcome the influence caused by the discontinuity of fuzzy parameters, but this method can only be limited to the case that the fuzzy process is established as a single-parameter Markov random model, and the calculation amount is also large.
Third, the application prospect of blind image restoration
(1) Improvement of existing algorithm and research of new algorithm. There are still many shortcomings in the existing algorithms, which need to be further improved. For example, in IBD algorithm, how to choose initial conditions to ensure the convergence of the algorithm; How to choose the termination condition of the algorithm to ensure the recovery quality; How to choose the noise parameters in the filter to reduce the influence of noise? For example, in NAR2IF algorithm, how to further solve the noise sensitivity problem, determine the support domain and how to extend the algorithm to the case of non-uniform background. A new algorithm is proposed to solve the problem of blind image restoration, which is also a hot research topic in the future.
(2) Blind image restoration algorithm based on nonlinear degradation model. In practical application, strictly speaking, all degradation models are nonlinear. The model is approximated by linearization method. Although the algorithm is simple, it is not ideal for dealing with serious nonlinear situations. The blind separation algorithm based on polynomial and neural network parameter model to deal with nonlinear signals needs further research when it is extended to two-dimensional images. The research of blind image restoration algorithm based on nonlinear degradation model is also one of the next research directions.
(3) Research on denoising algorithm. The existence of additive noise makes image restoration a morbid problem. Because it is generally assumed that only the statistical characteristics of noise are known, it is impossible to completely remove the noise in degraded images. In addition, because of the existence of noise, the restoration effect is not ideal, so it is of great practical significance to study the image blind restoration algorithm combined with noise reduction, and some work has been done in this regard. In order to overcome the influence of noise, noise is generally reduced first and then restored; Second, noise reduction and restoration are carried out at the same time. At present, most algorithms describe noise as Gaussian noise, which has great limitations in practical application. For non-Gaussian cases, it is also an important research direction to adopt denoising algorithm based on high-order statistical characteristics of noise. Other types of methods can also be used for denoising, and the nonlinear independent component analysis method of self-organizing mapping can be used for image denoising.
(4) Real-time processing algorithm. The complexity of the algorithm is an important aspect that restricts the application of the algorithm. Regularized discrete periodic Radon transform can be used to transform two-dimensional convolution into one-dimensional convolution for processing, thus improving the speed of the algorithm; You can also use the real-time processing algorithm of neural network. The real-time performance of the algorithm is the premise of its practical application.
(5) Applied research. The application of algorithm is the driving force to promote the research of algorithm. Although blind image restoration algorithm has been widely used in astronomy, medicine, remote sensing and other aspects, there is still a lot of work to be done to apply this algorithm to real-time detection of general industrial images, machine vision, image transmission restoration under network environment, criminal investigation and so on.
References:
Xue Mei, Yang Luxi. An improved NAS-RIF blind restoration algorithm for noisy binary images [J]. Data processing. 2006.4438+07.(2).
Tang Ting, Tao Qingchuan, He Xiaohai. Improved NAS-RIF blind image restoration algorithm based on wavelet denoising and image segmentation technology [J]. Journal of Chengdu University of Information Science and Technology. 2004.438+09。 (3).
Liu Cong, He Zhenya. Blind restoration of noisy blurred images based on singular value decomposition. [J]。 Data acquisition and processing 2002. 17. ( 1).