基于自然图像尺度不变特性的图像增强技术分析-analysis of image enhancement technology based on scale-invariant characteristics of natural images.docxVIP

本方法基于 1 中的第一点分析，由于不同尺度图像的响应分布越接近，那么该类导数滤波器越符合图像的尺度不变特性。那么，可以认为存在一个最 优的，即最符合尺度不变特性的，滤波器使得不同尺度图像的响应分布基本 相同。据此我们提出了基于直方图规定化的迭代方法，来学习该最优滤波器。 同时将该最优滤波器应用在图像的超分辨率处理方法中，该方法的主要思想 是根据最接近真实结果的响应来估计真实结果。关键词：尺度不变性；高斯混合模型(GMM)；滤波器；图像恢复；图像超分 辨率AbstractWith higher and higher Proportion of image and video in human communication, people have higher requirement for image quality, so image processing is getting more and more attention. To propose more efficient image processing algorithm, natural image statistics attract people’s attention, and are used in image processing. In the area of image processing these natural image statistics are used as constraints to improve processing image, and make the result closer to original, real image. Scale invariance is proposed as one of important statistic properties early, it is very effective in image processing. This paper analyses and researches natural image scale invariance, and the results are utilized in image restoration and image super-resolution. The main work and innovations are as follows:Analysis and model of natural image scale invariance.Firstly, natural image scale invariance shows that different scale images exhibit similar heavy-tailed distribution when derivative-like filters are applied to them. We think that the closer response distributions of different scale images are, the more derivative-like filters accord scale invariance.Secondly, for a natural image, it always contains the same contents of different scales and dually the same contents of same scale exist throughout scales of the image. According to it, we introduce Gaussian Mixture Model (GMM) to represent image information, and these information are used throughout scales.Scale invariant image super-resolution algorithmThis is a simple single image super-resolution algorithm. According to point 2 in section 1, GMM of small scale image is applied on large scale image patches processing across scale. We improve the results of super-resolution in this way, also it is