基于多信息融合的视觉目标类识别算法研究

视觉目标类识别是计算机视觉研究领域中的最具挑战性的难题之一,目前仍有许多问题没有得到很好的解决.近年来提出的空域金字塔直方图特征表示,在描述特征点集分布属性方面取得了比较好的实验效果.但是由于其描述的信息不全面,在性能上仍有较大改进余地.从信息互补性角度出发,提出了基于多信息融合的集成策略,将空域金字塔直方图表示与费舍分数表示各自描述的优势相结合,用于视觉目标类识别.实验证明该策略是有效的,在所进行测试的所有类别上相比单信息识别的性能均取得了一致性提高.

Multi-information for Visual Object Categorization

Visual object categorization(VOC) is one of the most difficult challenges in computer vision. Spatial pyramid histogram has been proposedd in recent years as an effective way to deal with features sets. However, there remains a large space for improvement. We made use of the respective advantage of spatial pyramid histogram and fisher score representation and proposed to use multi-information for recognition from information complement point view. The experiment results confirm our strategy, and our proposed algorithm consistently boosts the performance of all classes compared with their respective performances.