随着信息技术的持续进步,人们每天会接触到大量的图像,其形式和内容越来越多样化,色彩越来越丰富。可计算图像美感的研究目的是利用计算机模拟人类视觉系统与审美思维,对图像的美感程度做出判决。因此,图像美感质量评估研究逐渐引起研究者的兴趣,其研究结果可以应用到图像质量评估、摄影美感修正、艺术绘画分析等领域。目前,图像美感质量评估的关键问题有两个方面,包括如何获取表征能力强的图像美感价值信息特征,以及搭建设计美感评估模型。本文以研究图像美感自动评估问题为重点,实现图像的高低美感分类和评分任务。主要取得以下创新成果:1、提出一种融合图像场景信息的美感评估方法。新的方法针对图像的场景类型,综合考虑了不同类型图像的美感差异,使用一种新颖的卷积神经网络结构(Multi-scene Deep Learning Model,MSDLM),用于进行不同场景类型美感特征的学习。此外,为了减少训练数据量不足和噪声所造成的影响,对训练库做预处理的同时对模型增加了预训练步骤。从实验结果来,改进型的深度卷积神经网络获得了较好的分类效果。2、设计了一个综合的图像美感质量评估模型,该模型包括图像高低美感质量分类和美感评分预测两个任务。考虑到图像显著性区域的局部美感价值,设计并行双路卷积神经网络,一路用于图像全局美感特征的学习,一路用于图像显著主体局部美感特征的学习,最后融合得到美感预测结果。本文针对图像美感质量评估研究领域存在的问题,提出新的思路和模型设计,在主流的数据集上的评测取得了比其他方法更满意的效果,系统的输出结果与人类审美感知相符。
This paper presents a study on discovery learning of scientific concepts with the support of computer simulation. In particular, the paper will focus on the effect of the levels of guidance on students with a low degr...
详细信息
This paper presents a study on discovery learning of scientific concepts with the support of computer simulation. In particular, the paper will focus on the effect of the levels of guidance on students with a low degree of experience in informatics and educational technology. The first stage of this study was to identify the common misconceptions about density, starting from a literature review. Forty eight students (25 M and 23 F) from two high schools in Vanuatu were then involved in the study. These students were divided into three groups according to the different levels of guidance they received (Unguided;Minimum guidance, Maximum guidance). A pre and post activity questionnaire was designed containing 12 questions. The students underwent a training session with computer simulation about density. Using a descriptive and an inferential statistics method, scores obtained from the three different groups were compared during pre-test and post-test analyses. From the analyses it was found that the construction of knowledge from discovery learning activities occurs with or without guidance, however the amount of guidance received has an influence on the depth of conceptual understanding.
暂无评论