基于测地距离的超像素分析算法

超像素分析指的是将数字图像细分为多个超像素的过程,旨在简化或改变图像的表示形式,使得图像更容易理解和分析。文章提出了一种基于测地距离的超像素分析算法,该算法采用引入代价函数的Fast Marching算法来计算像素点间的测地距离。将目标图像大致均匀地划分成k个初始长方形区域,在每个区域内选取局部密度最大的像素点作为种子点,再由种子点出发计算像素点间的测地距离,并根据测地距离对像素点进行标记,故而可以得到大小均衡,形状规整的超像素。该算法在计算测地距离时,充分考虑了像素点的颜色和位置特征,并且以小区域为单位计算测地距离不仅缩小了Fast Marching算法的搜索范围,加快了算法的运行速度,还可以使得某些像素点的测地距离被重复计算,便于选取最优值。该算法所得超像素的分割精度及规整度都取得了良好的效果。

Superpixel Analysis Algorithm Based on Geodesic Distance

Superpixel analysis refers to the process of subdividing a digital image into multiple superpixels, which aims to simplify or change the representation of the image and make the image easier to understand and analyze. We propose a superpixel analysis algorithm based on geodesic distance, which uses Fast Marching algorithm with cost function to calculate the geodesic distance between pixels. The target image is roughly evenly divided into k initial rectangular regions. In each region, the pixel with the highest local density is selected as the seed point, and then the geodesic distance between the pixels is calculated from the seed point, and the pixels are marked according to the geodesic distance, so the size of the balanced and regular shape super-pixel can be obtained. When calculating the geodesic distance, the algorithm fully takes the color and location of pixels into account, and calculates the geodesic distance in a small area, which not only reduces the search scope of Fast Marching algorithm, speeds up the operation speed of the algorithm, but also makes the geodesic distance of some pixels be repeatedly calculated, which is convenient for selecting the optimal value. The segmentation accuracy and regularity of the superpixel obtained by this algorithm is effective.