Abstract: This paper presents a novel image super resolution method based on vector quantization.The idea is enlightened by the thought of fractal coding.The method uses learning algorithm to obtain the corresponding relationship between high frequency information and low frequency information of the single frame input image，and utilizes this relationship to add details at one octave above the spatial frequency in the input image in order to get the high resolution image.The method overcomes the shortcomings of traditional interpolation methods，which make image fuzzy because of smoothing images excessively，and preserve texture poor.The method is able to retrieve high frequency image details that are irretrievable when traditional interpolation methods are used.Experimental results show that this method performs better than traditional interpolation methods in terms of evaluations both objectively and subjectively.

Key words: image super resolution, vector quantization, fractal

摘要： 受分形编码思想启发，提出了一种新的基于向量量化的图像超分辨率方法。该方法使用学习算法来获取单幅输入图像中的高频信息和低频信息之间的对应关系，并利用此关系对输入图像的一个倍频程的空间频率内添加图像细节以获得高分辨率图像。该方法克服了传统插值方法中因过度平滑导致图像模糊和纹理保持较差的缺点，能够重现出传统插值方法不能复原出的一些高频图像细节。实验结果显示该算法在客观和主观上都比传统插值方法有更好的评价。

关键词: 图像超分辨率, 向量量化, 分形