Abstract: Aiming at the problem that the PhotoShop subjective examination questions can not realize the machine marking, it uses the method of image similarity comparison to conduct PS marking. The similarity degree is compared by extracting the feature histogram of the image, the transverse and the longitudinal similarity between the three images are defined, and the Euclidean distance normalization algorithm based on the distance triangle of similarity degree is proposed. This algorithm can calculate the similarity of the candidate is graph with respect to the color, edge, and texture of the original graph and the standard answer graph respectively, thus giving the actual score of the examinee. Experimental results show that the method is reasonable. In practical application, the weight of corresponding feature similarity can be adjusted according to the emphasis of examination on corresponding skills.

Key words: feature extraction, image similarity, Euclidean distance, automatic grading

摘要： 针对PhotoShop主观题无法实现机器阅卷的问题，运用图片相似度对比的方法来进行PS阅卷。通过提取图片特征直方图来进行相似度比对，定义了三幅图片之间的横向相似度与纵向相似度，提出基于相似度距离三角形的欧氏距离归一化算法。通过该算法计算出考生图相对于原图与标准答案图颜色，边缘，纹理的相似度，从而综合给出考生实际得分。实验结果表明，该方法所给出的考分合理。在实际应用中，可根据考试对相应技能考查的侧重，而调整相应特征相似度的权值。

关键词: 特征提取, 图像相似度, 欧氏距离, 自动评阅