基于改进SoftTriple的编程作业图像相似度量学习研究

Research of Image Similarity Measurement Learning Algorithm for Programming Coursework Based on Improved SoftTriple Model

目前,高校学生课程作业抄袭现象十分普遍,特别是编程相关作业中除了文字内容以外,还存在大量的图片抄袭。针对传统图片抄袭检测算法需要通过裁剪等方式固定输入图像的尺寸导致丢失图像信息问题,本文提出一种基于改进SoftTriple的图像相似性度量算法,用于高校课程作业中图片抄袭的检测。实验结果表明,本文提出的模型在Caltech101数据集上取得了最好的召回率和归一化互信息值。

Presently, the phenomenon of plagiarism in coursework has become very common, especially in programmingrelated homework. Apart from the text content, there are also many plagiarized images in coursework. The current algorithms need to fix the size of input images by cropping and other methods, which results in loss of image information. To address this issue, this paper proposed an image similarity measurement algorithm, which is used to detect image plagiarism in college coursework, based on an improved SoftTriple model. The experimental results show that the proposed model achieved the best recall rate and normalized mutual information value on the Caltech101 dataset.