中文摘要：

为了对内容感知图像缩放方法的性能及效果进行评价，提出一种新的主观评估方法．该方法主要包括3个方面的工作：首先提出了一种结合非完整型成对比较法及随机化区组设计的用户测评实验方案，采用非完整型成对比较法，不要求每个被测主体对实验数据集上的所有比较对进行比较，减少了因被测主体注意力不稳定而对测试产生的影响，通过随机化区组设计，尽量减少不同被测主体进行测试的比较对间的差异；然后实时检测被测主体的测评结果中是否存在循环误判，若出现循环误判，则返回相应的比较对让用户再次确认选择结果，以此减少由被测主体注意力不集中而引起的误判；最后结合多种非参数统计方法来分析实验数据，先检测多个被测主体对不同方法结果的选择是否有相同的倾向，在得出一致性倾向显著的基础上，结合Friedman检验、Page检验及Hollander-Wolfe比较来给出不同方法的效果排序及分组．针对多种典型内容感知图像缩放方法进行广泛的用户测评实验，结果表明，实验数据有较高的一致性系数，得出的主观测评结果被较多的被测主体支持．

英文摘要：

To effectively assess the property and efficacy of the numerous content-aware image resizing methods, a new subjective quality assessment method is presented in this paper. The method involves three part works. The first, the user test experiment scheme of combining the incomplete paired comparison and a randomized block design is proposed. Because the incomplete paired comparison design does not ask a participant to compare all pairs based on the experiment image set, this scheme can reduce the impact due to the test subject attention instability. Moreover, this scheme tries to minimize the discrepancy of the pairs compared by the different test subjects by the randomized block design. The second, the circle errors in the experiment results can be checked in real-time. Once a circle error is made, the test subject is asked to confirm his previous selection results to reduce the error results due to the thoughtless. The last, many non-parametric statistic methods are used when experiment data is analyzed. Specially, check the experiment data and see if the results selected by multiple test subjects on the same comparison pairs have same agreement preference. Once the agreement preference is salience, the effect of the different content-aware image resizing methods issorted and the methods are grouped by combining the Friedman test, the Page test and the Hollander- Wolfe comparison method. Based on the above experiment scheme and data analysis method, the large scale user study is conducted to assess a number of the representative content-aware image resizing methods. The superior coefficient of agreement from the experiment data shows that our experiment results are supported by the major test subjects.