基于改进机器学习的超分辨率图像细节复原

Rehabilitation of Ultra-Resolution Images Based on Improving Machine Learning

相对于低分辨率图像,高分辨率图像需要增加的像素数目更多,且需要增加高频信息以提升图像的清晰度,当图像目标与背景之间对比度较大时,图像高频细节信息复原难度较高。为此,提出基于改进机器学习的超分辨率图像细节复原方法。对图像去噪,并结合采用双边滤波方法实现图像的对比度增强;利用改进字典的机器学习算法建立双层字典,结合稀疏表示算法获取一层的粗略复原图像;通过二层字典计算一层复原图像与原始图像之间的差值,建立高分辨率样本,并对其开展二层字典训练,通过训练结构实现超分辨率图像的细节复原。实验结果表明,研究方法应用下峰值信噪比可保持在20dB以上,细节复原均方差低于4×10-3,结构相似性指标更高,高分辨率图像的训练效果更好,特征对比明显,细节信息突出。

Compared with the low-resolution image,the high-resolution image needs more pixels,as well as highfrequency information.When the contrast between the target and the background is high,it is difficult to restore the high-frequency details of the image.Therefore,a method of restoring super-resolution image details based on improved machine learning was proposed.Firstly,the noise was removed from images,and then the contrast of the image was enhanced by the bilateral filtering method.Secondly,the machine learning algorithm based on an improved dictionary was used to build a two-layer dictionary.Meanwhile,the sparse representation algorithm was adopted to obtain a rough restored image of the first layer.Furthermore,the difference between the restored image of the first layer and the original image was calculated through the two-layer dictionary.After that,high-resolution samples were established and trained by the two-layer dictionary.Finally,the detail restoration of super-resolution images was achieved through the training structure.Experimental results prove that the peak signal-to-noise ratio can be maintained above 20dB,and the mean square error of detail restoration is less than 4×10-3.In addition,the structural similarity index is higher,and the training effect of high-resolution images is better.The feature contrast is obvious,and the detail information is particularly prominent.