最小二乘匹配(LSM)对全局灰度和几何畸变有较好的适应性, 得到了广泛应用和重视。但是该方法存在计算量大, 难以实时应用, 以及不能适应局部光照变化等问题。提出了不变系数光照不变最小二乘匹配, 通过固定法方程式系数, 大幅减少了单位迭代计算量。在实际中, 由于成像场景和光照条件的复杂性, 传统的以整体灰度变换模型近似图像间灰度变化往往会存在较大误差, 由于真实灰度畸变模型不易获取和计算, 该问题一直没有得到很好解决。提出了用光照不变描述替换原始灰度描述的方法, 实现了对局部灰度畸变的适应, 并将最小二乘匹配的迭代维数由8维降为6维, 进一步提升了计算速度。理论分析和真实数据实验表明, 该方法较之传统方法计算速度提升6.18倍; 几何变化和局部光照变化仿真实验证明了此方法对局部光照变化有较好的适应性。

Least square match (LSM) can be resistant to global illumination and geometrical distortion. LSM, which is of widespread interest, has been used into many applications. But the method also has some problems, for example: huge computational cost, which makes it unrealistic for real-time applications and irresistibility to local illumination changes. To address the problems, a novel method namely illumination invariant least square match based on fixed coefficient is presented. Through improving the basic equations, the proposed method can largely reduce the computational cost of each iteration unit. In practice, global illumination transformation function usually has too much error when used to approximate the illumination changes between the matching images because of the complication of the scenarios and illumination conditions. This problem is still not solved properly, because the practical illumination transformation mode is quite difficult to model and calculate. The illumination invariant description is used to displace the traditional gray description. This scheme makes it resistant to local illumination change. In addition, the number of unknown parameters is reduced from 8 to 6, which further reduce the computational cost. The theory analysis and practical experiments demonstrate that the proposed method is faster 6.18 times than the traditional method. Simulation experiments using the local illumination changes and geometrical distortion images demonstrate that the proposed method can better handle local illumination changes.