Research On Color Image Super-resolution Algorithm Based On SVM Classified Learning

Due to the limitations of image capture device and imaging environments in traditional imaging process,high-resolution(HR)images are difficult to be obtained.Image super-resolution can use the method of digital image processing with one or an image sequence in original conditions to reconstruct HR images which over the range of imaging system.Reconstructed HR images gain high-quality performance in terms of both visual quality and computational cost such as Peak Signal-to-noise Ratio(PSNR)and Structural Similarity(SSIM).With the development of machine learning and pattern recognition,as an effective technique to deep learning,support vector machine has attracted considerable attentions.In this article,Image Super-resolution reconstruction technologies are studied and it comes from Basic and Frontier Technology Research Program Foundation of Tianjin(No.14JCQNJC00900): Research on Image Super-resolution Reconstruction using Support Vector Regression.One color low-resolution image is seen as the research object,and the specific work is as follows:Firstly,color image degradation mode was properly chosen to obtain low-resolution images according to characteristics of color low-resolution images.Secondly,in this paper an algorithm based on chromatic feature(HSV)was proposed.Support vector machine is used for learning and training.Before searching,according to the situation of small differences between objects and scenes in color low-resolution images,it selected the subset of sample library similar to the chromatic feature of object image to ensure the content relevance between the sample patch and the input low-resolution image.In addition,the algorithm reduced miss-matching times and reduced the computing complexity.HR images were reconstructed after the above procedure.The proposed algorithm was tested in the experiment and processing results were compared with the effects of two basic Super-resolution reconstruction algorithms.Compared with the bicubic interpolation method,the proposed method improves its Peak Signal-to-Noise Ratio(PSNR)and Structural Similarity Index Measurement(SSIM)by 3% ~ 5% and 2% ~ 4% on different images,respectively.This algorithm was proved to be an effective way to reconstruct low-resolution images.Finally,robust analysis was verified with the algorithm.For those low-resolution images which were added Gaussian noise below 2 standard deviation,the reconstruct results satisfied with the needs of the human eye and expected to achieve its objectives in Peak Signal-to-Noise Ratio(PSNR)and Structural Similarity Index Measurement(SSIM).