CAJ | 학술논문

图像窗口模式匹配前常需进行灰度分布标准化预处理，但由于某些模式识别场合需要掩去图像窗口的一些无关区域，常见的灰度分布标准化处理快速算法不适用。应用广义积分图像，实现了一种任意掩膜的图像窗口灰度分布标准化快速算法。掩膜的形状具有对称性时，算法的计算速度还可进一步提高。实验表明，该算法能提高掩膜图像窗口灰度均值、方差的计算速度1倍以上，可实现多尺度目标检测中对大量掩膜图像窗口的快速灰度分布标准化处理。
도상창구모식필배전상수진행회도분포표준화예처리，단유우모사모식식별장합수요엄거도상창구적일사무관구역，상견적회도분포표준화처리쾌속산법불괄용。응용엄의적분도상，실현료일충임의엄막적도상창구회도분포표준화쾌속산법。엄막적형상구유대칭성시，산법적계산속도환가진일보제고。실험표명，해산법능제고엄막도상창구회도균치、방차적계산속도1배이상，가실현다척도목표검측중대대량엄막도상창구적쾌속회도분포표준화처리。
It is often needed to normalize grayscale distribution of image windows before template matching,but sometimes a few regions of these windows should be eliminated by mask,normal rapid algorithms would be not suitable. As extended integral image is applied,an algorithm is presented for grayscale distribution normalization of image windows with mask of any shape. When the shape of mask is symmetric,the computing time can be further reduced. Experimental results indicate that the algorithm is twice or more as fast as the tradi-tional algorithm. It is especially suitable for object detection with multi resolution,when a large number of windows with mask are needed to be processed with grayscale distribution normalization.