基于相容粒度空间模型的自适应图像语义分类方法

Self-adaptive image semantic classification based on tolerance granular space model

针对图像底层特征和高层语义之间存在的语义鸿沟问题，运用相容粒度空间模型对图像语义分类进行了研究，提出一种自适应的图像语义分类方法，为解决此问题探索出了一种有效途径。该方法将图像集建模为基于原始特征的相容粒度空间；在此空间中，通过引入相容参数和构造距离函数来定义相容关系，从而通过调整相容参数可有效控制对象邻域粒的大小，最终可直接处理图像的实数型特征而无需进行离散化等预处理；此外，通过引入相容度的方法实现对相容参数的自适应优化，从而自动调整邻域粒的大小，使得构造的分类器几乎不需要手工设置参数即可自动适应于各种不同类型的图像集，并获得比同类算法更好的分类准确率。实验结果验证了这种方法的有效性和可行性。

Aiming at the problem of the semantic gap between the low-level teature and the high-level semantic, the pa- per uses the tolerance granular space model to study image semantic classification, and then proposes a self-adap- tive image semantic classification method, thus an effective way for solving the semantic gap problem is given. The proposed method models an image set as a primitive feature-based tolerance granular space, in which the tolerance relation is defined by using tolerance parameters and establishing a distance function, and then the size of an ob- ject＇ s neighborhood granule can be controlled effectively and finally the real-valued features can be directly dealt with without any pretreatment, such as discretization. In addition, tolerance parameters can be self-adaptively opti- mized by introducing the concept of tolerance degree, so as to automatically control the size of an object＇ s neigh- borhood granule, and in this way, the obtained classifier can adjust itself to a variety of image sets almost without any manual parameter configuration. The experimental results show that the proposed method is effective and feasi- ble, and it has better classification performance than that of similar methods.