摘要
随着人们对图像质量要求的不断提高,为了弥补原有图像数据空间分辨率的不足,重建质量更好、空间分辨率更高的图像数据,进一步提高图像空间解像力和清晰度,在对现有图像分辨率增强技术进行分析的基础上,提出了1种新颖的图像超分辨率算法—单向多样学习图像超分辨率算法。该算法思路是:首先建立训练图像集,再对待处理图像和训练集中的特征图像对进行分割、光栅排列和对比度正则化等适当的预处理。然后让待处理图像上的每个局部图像块在训练集中使用搜索算法进行多样学习,以获得低分辨图像上不同区域缺乏的高频细节信息,最后使用这些学习所获得的信息预测生成超分辨率图像。在本算法中,训练样本具有更好的一般化能力,训练集的通用性强,其图像存储空间减小,采用的最先最优搜索算法能快速找到了1个较好的匹配结果,从而以极小的计算代价提高了匹配质量,比采用图像超分辨率重建技术所获得图像质量更好。因此在军事遥感侦察、目标的识别与监测、天文观察、生物医学、公安侦破和交通监视等诸多领域具有广泛的应用前景。实验结果也表明,本文算法得到的高分辨率图像较大程度上提高了图像质量。
With the image quality requirements for the improvement of the original image data, in order to compensate for the lack of spatial resolution and reconstruct the quality of better, higher spatial resolution image data and further improve definition and image interpreting power of the image space. Based on the analysis of existing technology for image resolution enhancement, a novel algorithm for image super-resolution-one-way multiplicity learning image super-resolution algorithm was put forward. The algorithm is ： First of all, the training image set is established, and then some preprocessing operations are adopted in the object image and the character image of training image set, such as image segmentation, grating permutation, and regularization of contrast. Secondly, each local block of the object image on the use of search algorithm for multiplicity learning from the training set, thus some high frequency detail information, which are scarce for the low-resolution image, can be obtained. Lastly, these learning obtained high frequency detail information can be used to predict and produce the super-resolution image. In this algorithm, the training sample has better generalization ability, the training set has the characteristics of unified, the image storage space decreases, the use of the best-first search algorithm can quickly find a good match results, thus the minimal computational cost to improve the quality of the match, than the use of image super-resolution reconstruction of the image quality was better. In the fields of remote sensing in military reconnaissance, target identification and monitoring, astronomical observation, bio-medical, public security and traffic surveillance, it has a wide range of applications. The experimental results also show that the algorithm has been a greater degree of high-resolution images to improve the image quality.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2009年第7期953-960,共8页
Computers and Applied Chemistry
关键词
算法
图像
超分辨率
单向多样学习
algorithm, image, super-resolution, one-way multiplicity learning
