摘要： 提出了一种在复杂光照环境下具备旋转不变识别能力的匹配新算法。图像梯度方向角是一种在光照变化及目标发生旋转的情况下保持相对稳定的特征量。通过对该特征量的量化与统计分析，分2步实现在变化的光照条件下旋转目标的识别：(1)统计子图像块梯度方向角量化值的直方图，大概确定出目标旋转的角度；(2)利用确定的角度值对模板图像的梯度方向角量化值与子图像的量化值进行旋转匹配。实验证明该算法具有较强的鲁棒性。

关键词: 量化, 梯度方向, 直方图, 旋转不变, 光照变化

Abstract: A new rotation invariant matching algorithm which is robust against illumination variation is proposed. The gradient orientation angle is a relative stable feature in the condition of illumination fluctuations and object rotation. This paper uses a two-step method to realize the matching of rotated object in changed illuminations by quantifying and statistical analyzing this feature: (1)histograms of this feature of every subimage are employed for approximating the rotation angle of object; (2)rotated matching of templet and subimage in the space of quantification is performed by using the angle value. The effectiveness and robustness of this proposed algorithm has been proved by many experiments designed with real scene images.

Key words: quantification, gradient orientation, histogram, rotation invariance, illumination variation

中图分类号: 

• TP31