基于稀疏处理的图像质量评价方法研究

发布时间：2018-06-03 06:12

本文选题：客观图像质量评价 + 稀疏处理　；参考：《浙江大学》2016年硕士论文

【摘要】：在数字图像的获取、压缩、存储和传输过程中,由于存在获取设备的缺陷、压缩编码、存储错误和传输错误等问题,使得图片的质量下降,增加了人对图像内容识别的障碍。另一方面,在数字图像处理等领域中,图像质量的好坏直接代表了算法的性能优劣。而且图像质量作为一种重要的评价指标,可以用作优化图像处理系统参数的重要反馈。但是由于人对于数字图像的主观质量评价实验复杂,受实验环境影响大,且不可重复等各种因素,难以被大型应用场景采纳使用。客观图像质量评价作为一种预测数字图像质量,得出与人的主观评价结果相一致的机器算法,开始被研究人员广泛关注。稀疏处理是当前信号处理领域的研究热点,其优势体现在对信号进行高效表示,利用少量非零变量表征原始的大量数据,降低信号处理的复杂性。本文研究了稀疏处理的原理,并基于稀疏处理方法提出了自然图像质量评价方法,概括为:(1)调研了稀疏处理方法的基本原理,研究了核独立分量分析算法的原理和相关应用,利用核独立分量分析对数据进行非线性映射使得线性不可分的信号非线性可分的思想,设计了基于核独立分量分析的客观图像质量评价方法。方法对提取出的特征进行了基于自然图像统计学的统计分析,利用相关系数与图像质量产生映射关系。经过实验验证,核独立分量分析对图像数据进行分解得到的分量,作为有效特征可以较为精确的预测图像质量。(2)由于分量的独立性对于图像质量预测的精确性有直接影响,如何提取出更加独立有效的特征成为了基于核独立分量分析的图像质量评价方法的关键。使用频域距离作为块匹配方法的匹配要求,能够对图像采样数据进行高效的筛选。之后,经过主成分分析和核独立分量分析分解图像采样数据得到特征。经过实验证实,利用此种方法获得的特征经过质量综合,能较为明显的提升图像质量的预测精度。最后,论文总结了稀疏处理应用于图像质量评价框架中的思想和关键,说明稀疏处理对于自然图像质量评价的研究意义,总结了论文中基于稀疏处理方法中的分量分析的图像质量评价方法,对未来工作进行了展望。
[Abstract]:In the process of obtaining, compressing, storing and transmitting digital images, the quality of images decreases due to the defects of acquisition equipment, compression coding, storage errors and transmission errors, which increases the obstacle of image content recognition. On the other hand, in the field of digital image processing, image quality directly represents the performance of the algorithm. As an important evaluation index, image quality can be used as an important feedback to optimize the parameters of image processing system. However, the subjective quality evaluation of digital images is difficult to be used in large scale applications because of the complexity of the experiments, the influence of the experimental environment, and the non-repeatability of the experiments. Objective image quality evaluation, as a kind of machine algorithm which can predict digital image quality, comes up with the result of subjective evaluation, and has been paid more and more attention by researchers. Sparse processing is a hot research topic in the field of signal processing. Its advantage lies in the efficient representation of the signal, the use of a small number of non-zero variables to represent the original large amount of data, and the reduction of the complexity of signal processing. In this paper, the principle of sparse processing is studied, and the evaluation method of natural image quality based on sparse processing is proposed. The basic principle of sparse processing is investigated, and the principle and application of kernel independent component analysis (ICA) algorithm are studied. An objective image quality evaluation method based on kernel independent component analysis (KICA) is designed by using the idea of nonlinear mapping of data to make the linear inseparable signal nonlinearity separable by kernel independent component analysis (KICA). Methods the extracted features were statistically analyzed based on natural image statistics, and the mapping relationship between correlation coefficient and image quality was generated. It is proved by experiments that the components obtained from the decomposition of image data by kernel independent component analysis (ICA) can accurately predict the image quality as an effective feature) because the independence of the components has a direct impact on the accuracy of image quality prediction. How to extract more independent and effective features has become the key of image quality evaluation based on kernel independent component analysis (KICA). Using frequency domain distance as the matching requirement of block matching method, image sampling data can be filtered efficiently. After that, the image sampling data are decomposed by principal component analysis (PCA) and kernel independent component analysis (ICA). It is proved by experiments that the features obtained by this method can obviously improve the prediction accuracy of image quality through quality synthesis. Finally, the paper summarizes the ideas and key points of sparse processing applied in image quality evaluation framework, and explains the significance of sparse processing in natural image quality evaluation. In this paper, the image quality evaluation method based on component analysis in sparse processing is summarized, and the future work is prospected.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】