激光与光电子学进展, 2019, 56 (13): 131103, 网络出版: 2019-07-11
大样本图像质量主观评价方法 下载: 1101次
Subjective Image Quality Assessment for Large Samples
成像系统 图像质量评价 主观评价 模糊聚类分析 可察觉差异 图像质量数据库 人眼视觉特性 imaging systems image quality assessment subjective assessment fuzzy clustering analysis just-noticeable difference image quality databases human visual characteristics
摘要
针对图像质量数据库的主观评价方法存在失真等级少,缺少实验结果分析等问题,提出一种大样本图像质量主观评价方法。该方法基于双激励连续质量量表进行设计,使用简化的2级主观评价尺度评价,通过循环积分、最优选择和顺序调整获得样本图像的质量排序,并基于模糊聚类分析的思想将获得的图像次序的概率视为匹配程度,建立样本的模糊相似矩阵。通过指标规格化,建立模糊相似关系、等价关系以及分类、评分等步骤,最终确定图像质量得分。64级失真图像质量主观评价实验结果表明,图像质量得分能够准确反映可察觉差异的变化,主观评价结果的正确率达到94%,图像质量得分的标准差介于0~7,均值为3.08(百分制),远低于其他图像质量数据库的水平。所提方法具有较好的准确性和稳定性,适用于图像质量数据库的主观评价和人眼视觉特性研究。
Abstract
This study presents a novel subjective image quality assessment for large samples to solve existing problems in subjective assessments of image quality databases, such as less distortion levels and insufficient analysis of experimental results. The proposed method is based on a double-stimulus continuous quality scale and employs a simplified, two-level subjective assessment scale. We obtain a quality sequence of sample images by integrating circularly, selecting the best quality, and adjusting the sequence. Then, fuzzy clustering is used to analyze the quality sequence. The probability of image quality sequence in fuzzy clustering analysis is taken as its matching degree, which establishes a fuzzy similarity matrix of samples. We obtain the image quality score by normalizing the probability, establishing the fuzzy similarity relationship, and building a fuzzy equivalence relation, classification, and scoring. We test the subjective assessment for a 64-distortion-level image. The results demonstrate that the image quality scores accurately reflect the variation of just-noticeable difference, assessment accuracy is up to 94%, standard deviation of the image quality scores is from 0 to 7, and the mean value of standard deviation is 3.08 (percentile system), which is much less than the current level of other image quality databases. The proposed method demonstrates high accuracy and stability, and is suitable for subjective assessments of image quality databases and the study of human visual characteristics.
刘阳, 姜润强, 于洪君, 陈健. 大样本图像质量主观评价方法[J]. 激光与光电子学进展, 2019, 56(13): 131103. Yang Liu, Runqiang Jiang, Hongjun Yu, Jian Chen. Subjective Image Quality Assessment for Large Samples[J]. Laser & Optoelectronics Progress, 2019, 56(13): 131103.