采用亚像素缝雕刻的图像整形算法

摘要

全景拼接得到的图像其边界往往是不规则的，绝大多数全景拼接算法选择直接对全景图像进行裁剪，以获得拥有矩形边界的标准图像，但这会丢失部分图像中的关键信息并以减小全景图像的视野为代价。针对以上问题，提出一种基于亚像素缝雕刻的图像整形算法。首先采用三次样条插值和图像修复算法对图像的不规则边界进行平滑；然后以图像当前边界和矩形目标边界位置之差的归一化值为权重，在0~1个像素范围内拉伸或压缩最优缝上的每一个像素，平移缝一侧的其他像素，以消除当前边界和目标边界的差异。实验结果表明，在保证图像内容的前提下，该算法将一幅边界不规则的全景图像整形成一幅边界规则的矩形图像，在处理边界起伏较大，特别是有显著内凹边界的全景图像时也能取得令人满意的效果。%Stitched panoramic images mostly have irregular boundaries. In order to obtain standard images with rectangular boundaries, most panoramic stitching algorithm generally prefers to crop the images directly at the expense of decreasing the vision of panoramic images. An approach of rectangling images via sub-pixel seam carving has been proposed. It uses cubic spline interpolation and image completion to smooth the irregular boundaries. Taking the value of normalized difference between the smoothed boundaries and the expectable rec-tangular boundaries of the images as scale factor, it stretches or compresses each pixel on optimal seam in the range of 0-1 pixels and translates other pixels on one side of the seam repeatedly until the desirable boundaries are reached. Experimental results on a number of panoramic images show that the proposed method can achieve good results especially when the boundaries are very concave.