Abstract: An object-based tree representation method of the image (ORMG-tree) is presented, with multi-granularity in one scale, which combines improved polarity-based texture description with grey and position features, segments objects’ regions from the image with two domains partition method and the thought of minimal energy. The interested object regions and other regions with different granularity are represented, and the tree while segment the image is constructed. The experiment shows: the method emphasizes the representation and integrality of the interested object regions and ignores the background and uninterested regions, which makes the hierarchical image representation more effective and saves the processing time and storage space of tree constructing

Key words: Multi-granularity in one scale; ORMG-tree; Scale selection; Minimal energy; Image segmentation

摘要： 提出了一种基于对象的同层多粒度图像树型描述法(ORMG-tree)。它结合改进的基于极性的纹理表达，以及灰度、方位特征，采用两类区域法与能量最小化思想，从图像中分割出对象区域。通过不同粒度分别处理关注对象区域与非关注区域，并在使用ORMG-tree分割的同时，直接构造一个树型结构。实验表明：该方法充分突出了关注的对象性区域的表达及其完整性，尽量忽略背景与非关注区域，从而使得图像的多层表达更加有效，并节省了树型表达的处理时间和存储空间。

关键词: 同层多粒度；ORMG-tree；层选择方法；能量最小化；图像分割