一种基于图像区域系综分类的室外场景理解方法

多层感知机分类器是一种有效的数据分类方法，但其分类性能受训练样本空间的限制。通过多层感知机分类器系综提高室外场景理解中图像区域的分类性能，提出了一种自动识别室外场景图像中多种景物所属概念类别的方法。该方法首先提取图像分割区域的低层视觉特征，然后基于系综分类方法建立区域视觉特征和语义类别的对应关系，通过合并相同标注区域，确定图像中景物的高层语义。对包含5种景物的150幅图像进行测试，识别率达到了87％。与基于多层感知机方法的实验结果相比，本文提出的方法取得了更好的性能，这表明该方法适合于图像区域分类。此外，系综方法还可以推广到其他的分类问题。

An Outdoor Scene Understanding Method Based on Ensemble Classification of Image Regions

Even multi-layer perception (MLP) classifier has been an efficient method of data classification, and the performance is often limited by the training samples space. In this paper, the MLP classifiers ensemble is used to improve the performance of image region classification in understanding of outdoor scene and a scheme for automated recognition concept classes of objects in outdoor scene images by image region classification is presented. First, the low-level visual features are extracted from the segmented image region, and then the ensemble classifiers are used to establish corresponding relationship between the visual features of image region and semantic class. Finally, the high-level semantic class of each object in an image is formed by combining the region with same label. The method has been evaluated on 150 images including five objects and recognition rate is around 87%. The experimental results show that the proposed method that has better performance compared to MLP-based method is suitable for image regions classification. Moreover, this ensemble method appears to generalize to other classification problems. (