Research On Super-resolution Reconstruction Algorithm Of Face Image Via Dictionary With High Frequency Information

Image resolution is a major factor in limiting image quality.As images become more important in the field of information transmission,the demand of high-resolution images is increasing.Image super-resolution reconstruction technology can improve the resolution of the resulting image to a clear image by breaking through the limitations of image acquisition.Face image super-resolution technology known as face hallucination is a branch of the image super-resolution technology,specifically to face image processing to improve the recognition of fuzzy face images,now is widely used in face recognition and video surveillance and other fields.This paper mainly studies the super-resolution technique of face image based on sparse representation theory.Through the in-depth study of Yang's algorithm,the method of image sample library extraction is improved to improve the quality of reconstruction,and the efficiency of image reconstruction is improved by using step-by-step dictionary training method.The main achievements include the following two aspects:First of all,the image training sample library in the traditional reconstruction algorithm is established by random selection,which leads to the addition of redundant information to the training sample library and disturbs the image reconstruction result.At the same time,the random selection method makes the dictionary obtained during the image reconstruction process is unstable,resulting in unstable image reconstruction quality.Aiming at the above problem,this paper proposes a super-resolution reconstruction algorithm for face images based on dictionary with high frequency information.The high-resolution image information area is obtained by estimating the residuals of the high-resolution image obtained by enlarging the low-resolution image and the high-resolution.And then the image high-frequency information region is divided to establish the image sample training library.The dictionary by this training way has a better ability to express on the high-frequency information,so as to improve the quality of image reconstruction.In addition,selecting the high-frequency regional image block also cleverly solved the problem of image reconstruction instability.Through the experimental comparison,it is found that the reconstruction quality of the algorithm is slightly better than that of the traditional algorithm,and have a stable reconstruction result.Secondly,the traditional face image super-resolution reconstruction process using the joint dictionary on the training method to obtain high and low resolution dictionary taking longer.In order to improve the efficiency of image reconstruction,this paper proposes a super-resolution reconstruction algorithm for face image based on step-by-step construction of dictionary with high frequency information.In the training of high-frequency information dictionary on the basis of the use of high-resolution dictionary to solve the low-resolution dictionary to get the high and low resolution dictionaries.This dictionary training method reduces the training information dimension and improves the training efficiency.Simulation results show that this algorithm can effectively save the training time and only a small loss of image quality when the low magnification image is reconstructed.