Summary of sequential image super resolution restoration

 Summary: Super-resolution restoration of sequential images refers to the acquisition of one or more Sr restoration images by processing low-resolution degraded sequential images using signal processing methods, this technology can make up for the limitations of hardware implementation and reduce the cost. It has an important application prospect in video, remote sensing and other fields. This section briefly introduces the application of super resolution restoration, outlines the relevant main algorithms, and finally points out the development direction of this field.

Keywords: Super Resolution, image restoration, image processing

Abstract:Super-resolution image restoration technique is to estimate one or more super-resolution restoration images from a low resolution and degraed image quence via signal processing, which can feeds the limintion for hadware
Realization and costless. super-resolution restoration has applications, such as video, remote sensing imaging, and so on. the application of the technique is introduced, then main related algorithms are overviewed, and the direction for the future is
Given at last.

Key words:Super Resolution; image reconstruction; Image Processing

1. Introduction

Digital images collected by practical imaging systems (such as CCD, CMOS image sensors, and infrared imagers) are subject to the hardware implementation conditions and costs such as sensor arrangement density, and the resolution of the collected images is low, on the other hand, the imaging process is affected by many factors (such as optical system difference, atmospheric disturbance, motion, defocus, system noise, etc. [3]). this will also cause image quality degradation, such as blurring and deformation. Traditional image restoration technology can improve the quality of degraded images to a certain extent, but it does not change the resolution of restored images. In addition, high-resolution images have important applications in many fields, super-resolution (SR) is proposed in this context. The so-called Super Resolution restoration is to improve the quality of the acquired image while improving the resolution of the image through signal processing.

At first, the super resolution restoration technology only processes a single image. Its core idea is to improve the image resolution by estimating the high-frequency signal components beyond the cutoff frequency of the imaging system, this method has inherent limitations in image restoration because only one image is available. The super resolution Restoration Technology of sequential images is designed to use signal processing methods to obtain one or more Sr restoration images by processing low-resolution (LR) degraded sequences. Sequence Sr restoration can use additional information between frames, which is better than single Sr restoration. At present, it has become a research hotspot in this field.

This article briefly introduces the application of the sequence image Sr Restoration Technology, introduces the main implementation algorithms, and points out the development direction of this field.

2. Sr restoration application Overview

The super resolution image restoration technology can improve the resolution reduction caused by image discretization and degradation, make up for the deficiencies in the original image resolution, and break through the resolution limit of image acquisition methods, the potential of existing image data (such as multi-angle, multi-time, multi-platform remote sensing images, and sequential images) is explored. Therefore, the research on super-resolution image restoration technology is of great significance, it has important application values and broad application prospects in the following fields [7].

2. Video Processing

Currently, video images are evolving towards hd TV signals (HDTVs). Therefore, Sr technology is urgently needed to convert normal NTSC video images into hd TV signals. On the other hand, the human eye is not sensitive to the noise of the rapidly played video image due to the temporary effect of the human eye on the playback of normal video data. However, when a specific frame is printed, the effect will become very poor. Sequence Sr restoration technology can be used to obtain single-frame or multi-frame Sr images. Therefore, Sr restoration technology can be used to print video frames, this plays an important role in video capturing [7]. In addition, most of the current videos are compressed videos. The Sr Restoration Technology of the compressed videos plays an important role in improving the quality of the compressed videos [52] [53]. In addition, the SR restoration technology can work with mosaic processing and other video processing methods for video processing [54].

2.2 Pattern Recognition

Super-resolution image restoration can improve the quality of the collected images, provide more accurate image details, and thus improve the accuracy of pattern recognition, it plays an important role in improving the performance of some pattern recognition algorithms. For example, in the face recognition field, scholars have already carried out a lot of research work based on SR Image Restoration Technology, and have achieved good results. For more information, see [55]-[57]. in addition, Sr restoration technology can scale the areas of interest in the video surveillance [58] system to further improve the accuracy of monitoring target recognition, which plays an important role in traffic, Justice, and other aspects.

2. 3. satellite and remote sensing imaging

At present, there is a conflict between the amount of data in remote sensing images, that is, the amount of data obtained is large, but there is little useful information. The earth's resource satellites rotate around the earth about 14 times a day and cover the same area on the ground once in about 18 days. In this way, the data size of images that are repeatedly obtained in the same region is huge. However, due to the hardware technology of imaging equipment, images with higher resolution are often not obtained. At the same time, high-resolution cameras are expensive, the volume and weight are both large. Therefore, if we can use low-resolution camera imaging and perform super-resolution restoration of these duplicate image data to generate high-resolution images, we can reduce the risk and cost, at the same time, the application value of existing image data and the recognition capability of military targets are improved. In order to identify different environmental phenomena such as vegetation distribution and large body type, geological structure of the region, and regional scope of surface water, images obtained by satellites are multi-spectrum information including infrared, visible light, and ultraviolet rays. Therefore, image processing and analysis have become an important basis for data processing and analysis. Image processing visualizes resources, meteorology, and other information. Image analysis mainly uses Statistical Modes of multi-spectrum information to classify land use areas, meteorological, environmental pollution and resource investigation, and urban planning. Earth resource satellites can obtain multi-spectrum images. By processing these images in a series, different useful information can be obtained. However, due to the limitations of the existing imaging technology, the image resolution limits the identification and positioning accuracy of the image. To obtain high-quality and high-resolution images, it is impractical to simply increase the sampling rate and accuracy of physical devices. Even if possible, the cost of funds is expensive. Relatively speaking, it is not only feasible but also economical to obtain high-resolution images through multi-frame low-resolution image processing. The super-resolution image restoration technology uses existing remote sensing images to achieve the required recognition accuracy of the ground. Therefore, in the past, the field of image processing relying on optics is gradually becoming the mainstream in computer processing [7].

2. 4. Astronomy

To better understand and understand the universe, astronomers are trying to find more and better details in their astronomical images. By increasing the size of the telescope mirror or radio antenna, you can obtain better details or restore the collected images through post-processing methods. Under a limited budget, using post-processing technology is more practical than building expensive new equipment [7].

2. 5. Medical Imaging

In medicine, both basic medicine and clinical medicine require a large amount of medical image processing. The well-known X-ray images, microscope images, radioisotope images, ultrasonic images, magnetic resonance imaging (MRI) and other images have become the objects of Pattern Recognition in auxiliary diagnosis. Therefore, from the beginning of research in this field, image quality, precision, and image reconstruction have become one of the important objectives of Medical Image research. It is particularly widely used in chromosome analysis, automatic cell classification, identification of chest X-ray photos, treatment of fundus photos, and angiography analysis using fluorescent dyes. CT technology is a successful example of tomography reconstruction using multi-direction projection. Medical testing often needs to pass
CT technology identifies and determines the precise location and details of the disease (such as tumor), such as the shadow edge, the size and location of Foreign Body occupying. Due to the limitations of hardware devices and existing imaging technologies, high-definition images that meet high requirements cannot be obtained. Due to the special mechanism of CT technology, super resolution image restoration technology can be applied in this field. The use of super resolution technology to improve the quality and transformation of CT images and improve the image accuracy is still a topic currently being discussed in medical image processing [7]. In addition, magnetic resonance imaging (MRI) is also an important diagnostic method for clinical applications. It is of great significance to improve the resolution of the acquired images. [59], [60] for magnetic resonance imaging (MRI) the SR restoration technology is discussed in depth.

3. Sequence Sr Restoration Algorithm

The SR Restoration Algorithm Based on sequential images is more informative than the SR Restoration Algorithm Based on Single-frame images. The Restoration effect is obvious because the latter has become a mainstream research direction in this field. At present, the super resolution restoration of sequential images is mainly divided into two categories: Frequency Domain Method and airspace method. The observation model of the frequency-domain method is based on the shift feature of Fourier transformation. the advantages of the frequency domain method are: simple theory, low computational complexity, easy to implement parallel processing, and intuitive de-deformation super resolution mechanism; disadvantages: it can only be confined to global translation motion and linear space constant downgrading model, and has limited capabilities to include a prior knowledge of airspace. the observation model used by the airspace method involves global and local motion, spatial variable fuzzy point diffusion function, and non-ideal subsampling. It also has a strong ability to include the prior limitations of the airspace [14]. early work was mainly concentrated in the frequency domain. With more general research on imaging models, most of the later research was concentrated in the airspace [6]. Document [5] video sequence Sr restoration is divided into three categories: reconstruction-based, learning-based, and interpolation-based motion compensation. This article briefly introduces the frequency domain method of the SR Restoration Algorithm for sequential images, and then provides various main algorithms and other algorithms of the airspace method. For more information, see [1] ~. [6].

3.1. Frequency Domain Method

Tsay and Huang have made pioneering work on the Super Resolution Restoration Algorithm of sequential images. They first proposed an algorithm based on the frequency domain [8], the concept is to solve the spectrum of the original Sr by using the mixing relationship between the original Sr image and the DFT of multiple low LR images with global translation, and through the mixing spectrum of multiple LR images, then, use idft to reconstruct the original Sr image. This model is simple in theory and suitable for parallel processing. However, the observation model is limited to global displacement and LSI models. It is difficult to consider Spatial Constraints in the frequency domain [2] [4], therefore, the application scope is limited. Kim [9] ~ [12] tsay and other models have been improved and extended, but they have only improved the scope and speed of use, and there is no breakthrough progress. In addition, some scholars have also studied the DCT Domain [13] and Wavelet Domain [14] In other transform domains.

3.2. Airspace Law

Compared with the frequency domain method, the airspace method has better anterior constraints, in addition, the observation model can involve global and local motion, optical blur, intra-frame motion blur, spatial change point extended functions, non-ideal sampling, compression of pseudo images, and other content [7]. therefore, the application scope is wider. Elad and Feuer proposed a universal Observation Model of airspace SR-LR. this model describes translation rotation deformation, optical and motion blur, point extension function fuzzy, and undersampling in Image Acquisition. It provides good guidance for the design of the airspace algorithm. At present, Sr restoration technology is mainly used in airspace. common airspace methods include non-uniform interpolation, iterative Inverse Projection (IBP), and convex projection (POCS) the Maximum Posterior Estimation Method (MAP), maximum likelihood estimation (ml), and filter method among others. Among them, there are many researches on map and POCS, which makes great development space.

3.2.1 Non-uniform Interpolation

Non-All interpolation is intuitive and highly efficient. The algorithm flow consists of motion estimation (registration), nonlinear interpolation, Sr image fusion, and image restoration and deblurring. Here, the accuracy of motion estimation plays a key role in algorithm performance. Based on the generalized multi-channel sampling theorem [16] [17], ur and gross perform Nonlinear Interpolation on LR images with an overall spatial shift on the premise that the relative motion is accurate, then perform fuzzy processing [18]. Komatsu and others [19]. Use block matching technology to measure relative displacement and run landweber [20].
Algorithm, using multiple images taken by multiple cameras to obtain a SR image. Hardie [21] and others proposed a method combining gradient registration, weighted recent domain interpolation, and Vina Filter, FRANKE [22] and Sandwell [23] respectively proposed local thin Spline Interpolation and double harmonic spline interpolation SR interpolation algorithms.

3.2.2 iterative Inverse Projection (IBP)

Iterative Inverse Projection Method is intuitive and simple, but it is difficult to use a prior knowledge. Irani and releg proposed a variety of iterative Inverse Projection Methods [24]. The idea is to subtract the LR image simulated by the estimated Sr image from the observed actual LR image to produce an error, this process continues to iterate until the energy of the error reaches the minimum. Mann and PICARD [25] use a perspective motion model during image acquisition. Irani And Peleg [25] improved the IBP method and considered a more general motion model. Tom and others further improved the performance of iterative reverse projection algorithm by improving the motion compensation method, and applied the IBP algorithm [26] [27] in the SR restoration of the color video sequence.

3.2.3 convex projection method (POCS)

The POCS method is simple and can easily add prior information to any imaging model. The obtained high-resolution image edge and details are good, but its solution is not unique, the convergence speed is slow, and the stability is poor, the solution depends on the initial estimation and has a large computing capacity. POCS uses the idea of solving the intersection of space and a set of Constrained Convex Sets. A Convex Set can describe some prior properties of an image, such as positive definite, Bounded Energy, reliable data, and smooth data, this simplifies the solution space. POCS is an iterative process. When any point in the space of an ultra-resolution image is given, a point that satisfies the conditions of all convex attention sets can be located, that is, the convergence solution [7]. POCS was first proposed by stark and oskoui [28], tekalp [29]
Patti [30] [31] and others improved the POCS method, taking into account the Blur, sensor noise, motion blur, and noise problems caused by physical dimensions respectively. For other improvement documents, refer to [32]-[34].

3.2.4 Maximum Posterior Estimation (MAP) and maximum likelihood estimation (ML)

The Maximum Posterior Estimation (MAP) and maximum likelihood estimation (ML) are all restoration methods based on statistical theory under the Bayes framework. The maximum posterior probability estimation method regards super-resolution images as a map solution for complex optimization problems, and uses a prior smoothing hypothesis to reduce the influence of discontinuous measurements. The meaning of the Maximum Posterior Probability (MAP) is that the posterior probability of a high-resolution image reaches a maximum of [5] under the premise that a low-resolution video sequence is known. Schultz and Steven son use the Huber-Markov pair prior model to turn the problem into a constrained optimal problem with a unique solution. In the video sequence, a map
Sr restoration method [35]. Hardie et al. proposed a map algorithm that can carry out motion estimation and image restoration at the same time. It introduced a map cost function related to super-resolution images and registration parameters [36]. Giannis [37] et al. proposed a map-based SR restoration algorithm, which proposed a local boundary adaptive persistence strategy and then adopted an effective two-step reconstruction method, this method first registers low-resolution degraded images, and then uses the DFT domain descent Generation Algorithm for restoration, interpolation, and registration at the same time. Shen [38] et al. proposed a map-based algorithm for simultaneous motion estimation, motion segmentation, and Sr restoration.

The maximum likelihood estimation (ML) can be considered as a special case of the map method under the equiprobability prior model. Therefore, there is no prior term. Because Sr restoration is a disease issue, MAP has better performance than the ml algorithm. Tom and others use the EM algorithm to solve the ML estimation problem, which can simultaneously estimate the variation of sub-pixel displacement and noise [39] [4].

3.2.5 Filter Method

The filtering method is mainly used in scenarios with high real-time requirements such as video processing. It pays more attention to Algorithm Execution speed and has lower requirements on SR restoration performance than map and POCS. Elad [40] [41] et al. proposed a least square estimator based on adaptive filtering theory. Alam [42] et al. introduced the Gini filter method, Elad
[43] [44] et al. proposed two iterative algorithms of recursive steep descent algorithm (R-SD) and recursive least mean square algorithm (R-LMS, these two methods can be considered as the approximation of the Kalman Filter [7]. Narayanan [46] et al. proposed the SR Restoration Algorithm of The PWS filter (partition-based weighted sum filters.

3.4. Learning Algorithms

Document [46] proposes a SR restoration algorithm based on prior knowledge recognition, that is, a learning algorithm that learns to recognize specified categories, such as objects, scenes, and images, the obtained recognition prior knowledge is used for super resolution restoration. Capel [47] uses an image learning model in the Bayesian framework, which is much higher than the traditional ML estimation in terms of Super Resolution enhancement. Baker [48] et al. proposed a hallucination algorithm that integrates low-resolution feature recognition, freeman [49] et al. proposed a new method for generating high-frequency details by interpolation from the training set. Learning-based algorithms are a novel method of traditional Sr restoration algorithms, which can solve many difficulties in traditional methods. However, there are still many imperfections in the process of modifying algorithms, it is worth further research.

4. research prospects

Super Resolution Restoration Technology is a cutting-edge hot research area in the digital image processing field. It has important application values and potential in many fields, this article briefly introduces the application and main implementation algorithms of the super resolution restoration technology. At last, we will provide further research directions in this field. The literature [1]-[6] has been discussed and summarized as follows:

(1) more accurate motion estimation algorithms. In Sr restoration of sequential images, the accuracy of motion estimation (Registration) has a decisive impact on the performance of the entire algorithm. Although there are already many mature motion estimation algorithms, but the results are not very good. Therefore, developing motion estimation algorithms with better performance is an important and fundamental research direction for future Sr restoration technologies.

(2) Developing and proposing a new degraded imaging model is a more universal model in use and more accurate estimation of point diffusion functions and noise.

(3) Blind Sr restoration technology. In a large number of SR restoration algorithms, it is assumed that the degradation model is known, but in many practical scenarios, this premise is not satisfied. Therefore, it is necessary to study the blind Sr restoration algorithm when the degradation model is unknown.

(4) Compression Domain Sr restoration technology. Because the current video is mostly compressed format, it is necessary to carry out directly in the Compression Domain (especially the latest MPEG-4, H. 264 and other new video coding standards) to carry out Sr restoration research to improve algorithm efficiency.

(5) Research on MAP/cops fusion. These two algorithms have the most research in SR Restoration Technology, each with its own advantages and disadvantages. Therefore, the fusion and complementary research of these two mainstream algorithms can generate more effective new algorithms. In addition, other existing algorithms can be further studied to improve and integrate with each other.

The above points out that they are the main development directions of SR restoration technology in the future. Here, I will give you some ideas.

(1) Conduct in-depth research on specific applications of SR restoration technology. Sr technology has broad application prospects, but different application backgrounds have special implementation characteristics of algorithms. Therefore, while developing universal algorithms, it is necessary to develop specific algorithms for specific application environments, it can better realize the Running Effect of algorithms. In addition, it is also an important research direction to further expand the application field of SR restoration technology.

(2) Conduct in-depth research on SR Restoration Technology and other fields. In the document [38], motion estimation, motion segmentation, and Sr restoration are carried out at the same time. A preliminary exploration is made on the integration of SR demobilization and other research fields, this idea can be extended to other related fields. For example, the sequence Sr restoration technology can be combined with the dynamic visual attention model [50], Video Object Segmentation, and tracking [51] research, only for areas of interest (ROI) in the video sequence) or perform Sr restoration for the moving object.

(3) conduct in-depth research on real-time Sr restoration algorithms and related hardware implementations.

References

[1] Borman S, Steven son R. spatial Resolution enhancement of low-resolution image sequences: a comprehensive review with ctions for future research [R]. lab. image and Signal Analysis, University of Notre Dame, Tech. rep ., 1998.

[2] Park s c, Park M k, Kang m g. super-resolution image reconstruction: a technical review [J]. IEEE Signal Processing magazine, 2003, (5): 21-36

[3] su binghua, Jin Weiqi, Niu Lihong, and others, super-resolution image restoration and progress, optical technology, (1): 6-9.

[4] research progress of Super Resolution Image Restoration Technology, measurement and control technology, (5): 1-5

[5] Wang Yong, Zheng Hui, Hu Devin, video Super Resolution enhancement technology overview, Computer Application Research,: 4-

[6] Zhou Fang, Super Resolution Image Restoration and progress, automation and instrumentation,-14

[7] Guo xiaoxin, research on the problem of super resolution reconstruction, Ph.D. thesis of engineering, Jilin University, 2005.

[8] Tsai r y and Huang t s. multipleframe image restoration and registration [A]. advances in computer vision and image processing. greenwich, CT: JAI Press Inc ., 1984: 317-339.

[9] Kim s p, Bose n k, and Valenzuela H m. recursive reconstruction of high resolution image from noisy undersampled multiframes [J]. IEEE Trans. ACO ., speech, signal processing, 1990,38 (6): 1 013-1 027.

[10] Kim s p and Su w y. Recursive High-Resolution Reconstruction of blurred multiframe images [J]. IEEE Trans. Image Processing, 1993,2 (10): 534-539.

[11] Bose n k, Kim h c, and Valenzuela H m. recursive Implementation of total least squares algorithm for Image Reconstruction from noisy, undersampled multiframes [A]. in Proc. IEEE Conf. acoustics, speech and signal processing, minneapois,
Mn, 1993,5: 269-272.

[12] Rhee s h and Kang m g. discrete Cosine Transform Based regularized high-resolution image reconstruction algorithm [J]. OPT. eng ., 348-1 356.

[13] s.h. rhee and M. g. kang, discrete cosine transform based regularized high-resolution image reconstruction algorithm, OPT. eng ., vol. 38, no. 8, pp. 1348-1356, Aug. 1999.

[14] Zhang xinming and Shen lanyun, achieving super-resolution restoration of images in the wavelet transform domain, Journal of computer science, 1183 (9): 118-

[15] M. elad and. feuer, restoration of a single Superresolution image from several blurred, noisy, and undersampled measured images, IEEE Trans. image processing, vol. 6, No. 12, pp. 1646-1658, Dec. 1997.

[16] A. Papoulis, generalized sampling theorem, IEEE Trans. Circuits syst. vol. 24, pp. 652-654, nov. 1977.

[17] J. L. Brown, multi-channel sampling of low pass signals, IEEE Trans. Circuits syst., Vol. CAS-28, pp. 101-106, Feb. 1981.

[18] H. Ur and D. Gross, improved resolution from sub-pixel shifted pictures, cvgip: graphical models and image processing, vol. 54, pp. 181-186, mar.1992

[19] T. komatsu, K. aizawa, T. igarashi, and T. saito, signal-processing based method for acquiring very high resolution image with multiple cameras and its theoretical analysis, Proc. inst. elec. eng ., vol. 140, No. 1, PT. i, pp. 19-25, Feb.
1993.

[20] L. landweber, an iteration formula for fredholly integral equations of the First Kind, Amer. J. Math. vol. 73, pp. 615-624,195 1.

[21] M.S. alam, J. g. bognar, r.c. hardie, and B .j. yasuda, infrared image registration and high-resolution reconstruction using multiple translationally shifted aliased video frames, IEEE Trans. using um. MEAs ., vol. 49, pp. 915-923, Oct. 2000.

[22] Richard Franke. smooth interpolation of scattered data by local Thin Plate splines. comput. Math. Appl. 1982, 8 (4): 273-281

[23] David T. Sandwell. biharnionic Spline Interpolation of Geo-3 and Seasat altimeter data. Geophysical Research Letters. February 1987, 14 (2): 139-142

[24] M. Irani, S. Peleg. Improving Resolution by image registration. cvgip: graphical models and image Proc., May 1991, 53: 231-239

[25] S. Mann, R. W. Picard. Virtual bellows: Constructing High Quality stills from video. In Proc.

IEEE Int. conf. Image Processing, Austin, TX, nov. 1994. 363-367.

[26] B. c. tom,. k. katsaggelos. resolution enhancement of video sequences using motion compensation. in: Proc IEEE int conf image processing. lausanne, Switzerland. 713, 716 :-

[27] B .C. Tom, A. K. katsaggelos. Resolution enhancement of color video. In: Proc conf EUR signal processing. 1996.-148

[28] Stark H and oskoui p. high Resolution Image Recovery from image-plane arrays, using convex projections [J]. j. OPT. soc. am. a, 1989,6: 1 715-1 726.

[29]. m. tekalp, M. k. ozkan, M. i. sezan. high-resolution image reconstruction fromlower-resolution image sequences and space varying image restoration. in Proc. IEEE Int. conf. acoustics, speech and signal processing (icassp), San Francisco,
CA., Mar. 1992,3: 169-172.

[30] J. Patti, M. sezan, A. M. tekalp. High-resolution image reconstruction from a low-resolution

Image Sequence in the presence of time-varying motion blur. In: Proc IEEE int conf image processing. Austin, TX. 1994. 343-347

[31] J. patti, M, I. sezan,. m. tekalp. high-resolution standards conversion of low resolution video. in: Proc IEEE int conf ASSP. detroit, MI. 1995. 2197-2200

[32] Tom B C and katsaggelos a K. an Iterative Algorithm for improving the resolution of video sequences [A]. in Proc. 1996 Spie Conf. visual communications and image processing, Orlando, FL, 1996: 1 430-1 438.

[33] Hong m c, Kang m g and katsaggelos a K. A regularized multichannel estoration approach for Globally Optimal High Resolution video sequence [A]. in Spie vcip, vol. 3024, San Jose, CA, 1997: 1 306-1 317.

[34] p. e. eren, M. i. sezan,. m. tekalp. robust, Object-based high-resolution image reconstruction from low-resolution video. IEEE Transactions on image processing, Oct. 1997,6 (10): 1446-1451

[35] Schulz R and Steven son r l. Extraction of high-resolution frames from video sequences [J]. IEEE Trans. Image Processing, (6): 996-1 011.

[36] r. c. hardie, K. j. barnard, E. e. armstrong. joint map registration and high-resolution image estimation using a sequence of undersampled images. IEEE Transactions on image processing ., dec. 1997, 6: 1621-1633

[37] giannis K. chantas, Nikolaos p. galatsanos, and Nathan. woods, super-resolution based on fast registration and maximum a posteriori reconstruction, IEEE Trans. image processing, 2007,16 (7): 1821-1830

[38] huanfeng Shen, liangpei Zhang, Bo Huang, and Pingxiang Li, a map approach for joint motion estimation, segmentation, and super resolution, IEEE Trans. image processing, 2007,16 (2): 479-490

[39] B. c. tom,. k. katsaggelos. reconstruction of a high-resolution image by simultaneous registration, restoration, and interpolation of low-resolution images. proc. 1995 IEEE Int. conf. image processing, Washington, DC, Oct. 1995. -542.

[40] M. Elad, A. Feuer. Superresolution restoration of an image sequence: adaptive filtering approach. IEEE Transactions on image processing, Mar. 387: 395-

[41] M. elad,. feuer. super-resolution restoration of continuous image sequence using the LMS algorithm. in: Proc 18th IEEE Conf Israel, Tel-Aviv. 1995, 2.2.5: 1-5

[42] M. s. alam, J. g. bognar, R. c. hardie, B. j. yasuda. infrared Image Registration and high-resolution reconstruction using multiple translationally shifted aliased video frames. IEEE Transactions on capacity um. MEAs ., oct. 2000, 49: 915-923

[43] M. elad,. feuer. super-resolution reconstruction of image sequences. IEEE Transactions on Pattern Anal. machine Intelli ., sept. 1999, 21, (9): 817-834

[44] M. Elad, A. Feuer. Super-resolution reconstruction of Continuous Image Sequences. In: Proc int conf image Processing-1C1P1999. Kobe, Japan. 1999. 459-463

[45] Balaji Narayanan, Russell C. Hardie, Senior

Kenth E. barner, and Min shaoa computationally efficient super-resolution algorithm for Video Processing Using partition filters, IEEE transaction on circuits and system for video technology, 2007,17 (5): 621-

634

[46] Baker S, kanade T. Limits on super2resolution and how to break them [J]. IEEE Transactions Pattern Analysis and machine intelligence, 2002, 24 (9): 1167-1183

[47] Capel D, Andrew zisserman. super2resolution from multiple views using learnt image models [Eb/Ol]. http: // www. robots. ox. AC. UK/vgg/publications/html /.. /papers/capel2001. ps. GZ, 2002.

[48] Baker S, kanade T. super2resolution: reconstruction or recognition [C]. ieee2eurasip Workshop on nonlinear signal and image processing, Baltimore, Maryland, 2001,215-219.

[49] Freeman W, et al. Example-based super-resolution [J]. IEEE Computer Graphics and applications, 2002, 22 (2): 56-65.

[50] María T. lópez, Miguel. fernández, Antonio Fernández-Caballero, Jos é Mira and Ana E. delgado, Dynamic Visual Attention model in image sequences. image and vision computing. 22 (5), May 1, 2007, pp. 597-613.

[51] Yi Liu, Yuan F. Zheng, Video Object Segmentation and tracking using classification-learning classification, IEEE transaction on circuits and system for video technology 885 (7): 899-

[52] Y. altunbasak,. patti, R. mersereau, super-resolution still and video reconstruction from MPEG-coded video, IEEE Trans. circuits systems video technology 12 (2002) 217-226.

[53] B. gunturk, Y. altunbasak, R. mersereau, multiframe Resolution enhancement methods for compressed video, IEEE signal Proc. Let. 9 (2002) 170-174.

[54] P. Kramer, Et Al., super-resolution mosaicing from MPEG compressed video, signal process. Image

Commun. (2007), doi: 10.1016/j. image.2007.06.004

[55] gunturk, et al, eigenface-Domain Super-resolution for face recognition, IEEE Trans. Image Processing, 2007,12 (5): 597-606

[56] K. jia, S. gong, multi-modal tensor face for simultaneous super-resolution and recognition, Proceedings of the tenth IEEE International Conference on Computer Vision (iccv '05), 2005

[57] Jiangang Yu and Bir Bhanu, super-resolution restoration of facial images in video, the 18th International Conference on Pattern Recognition (ICPR '06), 2006

[58] gulcin canel, A. Murat tekalp And Wendi heinzelman. Super Resolution recovery for multi-camera surveillance imaging, proceedings. 2003 International Conference on multimedia and Expo (ICME '03), 2003

[59] H. Greenspan, Etc., MRI inter-slice reconstruction using super-resolution, magnetic resonance imaging, 20 (2002): 437-446

[60] Eyal Carmi, etc. Resolution enhancement in MRI, magnetic resonance imaging 24 (2006): 133-154