摘要： 基于区域的立体匹配是宽基线立体匹配的主要方法，其难点在于不变区域的提取。该文提出了稳定极值区域的概念，建立了提取该区域的算法。该算法应用了灰度指数变换，由I空间变换到 空间，以突显灰度值的变化，易于鉴定区域的边界。在此基础上计算了区域的仿射不变矩及衡量对应区域的欧氏距离。实验表明，该方法所提取的极值点定位准确，区域稳定性好，对应区域精度提高到70.9%。

关键词: 极值点, 指数变换, 稳定极值区域, 仿射不变矩

Abstract: Stereo matching with regions is a key method for binocular image matching of wide baseline and it is difficult to extract the invariant region. This paper proposes stable extremal region(SER) and its extracting algorithm. To magnify the intensity change and identify its edges easily, the exponent transform is used. It computes the affine invariant moments of SER and measures the Euclid-distance between the region and another. Experiment result shows that the extremal points locate accurately and the SER are stable. The corresponding regions precision improves to 70.9%.

Key words: extremal point, exponent transform, stable extremal region(SER), affine invariant moment

中图分类号: 

• TP391.41