区域选取与差异图融合的目标变化检测方法

针对现有目标变化检测方法用于高分辨率可见光图像时,未能有效利用图像的局部特征,目标误检率高。提出了一种区域选取的目标变化检测方法。首先利用整体变分算法获取两幅源图像的结构图,并将两幅源图像与两幅结构图进行分块;然后通过计算得到两幅结构图相同位置块的熵值差和两幅源图像相同位置块的结构相似度,并将两者分别与各自阈值比较以确定出存在变化目标的图像块;最后将存在变化目标的图像块的对数比差异图与均值比差异图进行融合,并对融合后的图像进行形态学处理,得到最终的目标变化检测结果。结果显示,与阴影补偿方法相比,算法的目标目标漏检率降低了36%,伪目标率降低36%,数据表明该方法能够准确的检测出真实变化的目标,较好地排除了伪目标。

Change Detection Based on Regional Selection and Difference Images Fusion

Aiming at Existing methods of high resolution visible light image change detection direct detection of the whole image. The local features of the image can not be used effectively and the detection results have the false target of the shadow change. A method of regional selection is proposed. Firstly, using the method of total variation calculate the structure images of two source images. Then block two pieces of the source image and two pieces of structure image. Next calculate the entropy difference of same position image block of structure image, and structural similarity of same position image block of source image. Compare entropy difference and structural similarity with respective threshold. Identify the image block that has the change target. Last, fusion log- ratio difference and mean- ratio difference of the image block that has the change target. Morphological processing of the fused images. Results shows that, Compared with the shadow compensation method, the algorithm of the target pixel detection rate is reduced by 34%,False target rate decreased by 36%. The data shows that the method can detect the real target accurately and eliminate the false target.