基于积分图像的快速模板匹配

Rapid Algorithm of Template Matching Based on Integral Image

人脸检测中,模板匹配前往往需要对图像窗口进行灰度分布标准化,而灰度分布标准化则要先算出图像窗口的灰度均值及方差。在积分图像基础上,提出和应用平方积分图像,实现了一种计算速度与图像窗口大小无关的快速灰度均值和方差算法。应用这种快速算法,结合对灰度分布标准化近似计算公式、相关系数和平均偏差计算公式的变换,以及灰度分布标准化处理时省略拷贝图像窗口及窗口灰度值变换等不必要的步骤,大大提高了模板匹配速度。人脸检测实验证明这种快速模板匹配算法是有效的。

In face detection, it is needed to normalize grayscale distribution of image windows before template matching, and that the grayscale average and variance must be computed before grayscale distribution normalization. Based on integral image, the square integral image is proposed and used to realize a rapid algorithm for calculating grayseale average and variance of image windows, and the calculating speed is independent of the windows＇ size. With this rapid algorithm and transformation of some formulas, such as approximately formulas for computing grayscale distribution normalization, formulas for computing the correlation coefficient and average deviation, and leaving out some needless processes such as copying windows and conversion of windows＇ grayscale value, the template matching is sped up. Experimental results of face detection demonstrate the efficiency of this rapid algorithm for template matching.