基于BoF模型的图像表示方法研究

Study of BoF Model Based Image Representation

设计合适的图像表示是计算机视觉中最重要的问题之一。BoF特征表示方法非常流行,已经广泛应用于图像分类、对象识别、图像检索、机器人定位和纹理识别。BoF特征是将图像表示为无序的特征集合。这种方法虽然缺乏结构信息和空间信息,但概念简洁、计算简单,在某些应用上取得的效果甚至可以与当前最好的方法媲美。仔细研究了BoF模型,着重对BoF模型中的3个阶段:局部特征提取、特征量化和编码、特征汇集所涉及到的典型技术进行了讨论。最后在分析各类研究方法的基础上,总结了目前研究存在的问题及可能的发展方向。

Designing a suitable image representation is one of the most fundamental issues of computer vision. BoF model is very popular and used extensively in image classification,video search, robot localization and texture recognition. BoF feature is an orderless collection of quantized local image descriptors. While this feature representation discards structural and spatial information,B oF model is conceptually and computationally simple, even as good as stateof the- art methods. Three steps in the popular BoF were studied in detail, including feature extraction, feature coding and fea- ture pooling. In the end, the main problems and challenges were highlighted based on analysis of current research tech- nique.