C-SIFT特征结合空间金字塔描述的情感图像分类

Emotional Image Classification Based on Color Scale Invariant Feature Transform Feature and Spatial Pyramid Model

摘要: 情感图像分类的目的是希望计算机能够表述人类观察图像时所引起的情感反应，并根据这种反应把图像分到不同的情感类别。本文提出了一种基于空间金字塔的情感图像分类方法，首先对图像提取颜色尺度旋转不变特征，并聚类形成视觉特征词典；其次对图像进行空间金字塔分块，使用局部约束线性编码方法表示各子块图像，形成图像的空间金字塔描述；最后通过训练分类器实现对情感图像的分类。该方法在SIFT的基础上加入了具有表征感情色彩的颜色特征，提取了独特的图像情感特征。本文方法分别在国际情绪图片系统(IAPS)数据库和人脸情感数据库(KDEF)上进行了实验，取得了较为理想的情感分类结果。

Abstract: The purpose of emotional image classification is that the computer can express the emotion reaction when observing the image, and classify the images into the different emotional categories automatically. In this paper, we proposed an emotional classification framework based on the spatial pyramid representation. First, we extracted the SIFT (Scale Invariant Feature Transform) feature of the colour, and performed the clustering method to form the codebook. Then, each image is described by using LLC (Local-constrained Linear Coding) scheme, and image representations were performed by the methodology of spatial pyramid. Finally, we performed the emotional categorization by the training classifier. As the colour information is significant to human visual perception we added an extra colour feature. In the experiments made on the IAPS (International Affective Picture System) and KDEF (Karolinska Directed Emotional Faces), an ideal classification result was obtained.