由于采用普通照相机拍摄的照片可能出现图像扭曲、交叠和倾斜,现有的图像镶嵌算法仍不能有效的进行处理,而且在这方面的研究也相对甚少。本文针对此问题,提出了一种适用于图像发生任意角度的旋转变换下仍可实现精确镶嵌的算法。即首先是采用Harris算子进行角点提取,使提取的精度达到亚像素级;然后根据这些角点信息建立图像间的角点对应关系,并由此得到投影变换矩阵;接下来采用图像变形技术,通过逆向映射重建发生空间变换的图像,由此得到的变形图像与源图像再进行无缝拼合。实验证明,该算法是有效的,具有较高的镶嵌精度。
So far, many scholars have put forward a lot of the mosaicing methods of obtaining panorama images, which only were the same with some application domain because of their localization. So a method of image mosaic based on comer detection is presented in the case of image rotating random angle. Firstly, Harris operator is used to extract feature points which can obtain sub-pixel precision. Then the matching comer pairs among these frames can be found. Based on these pairs, the matrix of projection transformation can be gained. Following this, it uses the technique of image metamorphosis with inverse mapping to resample and interpolate for the new image. Finally, it synthesizes novel view from the interpolated image and the source image and eliminates seam-lines. The experiments show that the method can obtain good mosaicing results with high stability and precision.