一种双层情感图像检索模型(英文)

A Double-Level Emotion Image Retrieval Model

随着信息技术的迅猛发展,情感信息处理已成为21世纪人工智能领域所面临的重要挑战之一。借鉴认知心理学、绘画艺术和服装设计的研究成果,本文提出了一种双层情感图像检索模型。在该模型中,借鉴心理学中的“维量”思想,建立情感空间;同时,抽取图像中较容易引起情感变化的特征作为图像的视觉特征,建立图像的特征空间;另外,本文还提出了情感注释的思想,采用支持向量机的方法建立图像的低层特征空间到用户的高层情感空间之间的映射,自动注释用户未曾评估的图像,实现了图像情感注释,在情感空间进行公共情感检索,快速获得用户情感信息,在此基础上,采用可视化交互式遗传算法实现因人而异的个性化情感检索,该模型应用于风景和服装图像的情感检索,取得了较好的实验结果。

With the rapid development of information technology, emotion information processing has become a great challenge faced by researchers of Artificial Intelligence (AI) in the 21st century. Inspired from the research products and methods of cognitive psychology, painting and design, a double-level emotion image retrieval model that consists of common emotion retrieval and individual emotion retrieval has been proposed in this paper. First, based on the idea of “dimension” from psychology, an orthogonal common emotion space is constructed. Then, sensitive features are extracted from images to construct the feature space. After that, emotion annotation idea has been provided, support vector machines are used to map images from the low level feature space to the high level emotion space, and automatically annotate unevaluated images based on users’ common emotion. Thus common emotion image retrieval is implemented by indexing images in the common emotion space and user’s subjectivity inherent in vision can be quickly grasped. Furthermore, an interactive individual emotion image retrieval using visualized interactive genetic algorithm is presented to adapt to individual variation and improve accuracy of the retrieval results. Last, an emotion scenery and fashion image retrieval system has been realized. The experimental results demonstrate the effectiveness of our approach.