天天向上

今天在google有關shape matching的主題, 意外搜尋到OpenCV3.0.0.0的線上資料

我想關於OpenCV 3.0將提供<ShapeDistanceExtractor>的介面, 應該是補足目前OpenCV 2.4.6 所提供的cv::matchShapes

另外, 外觀比對還有其他著名的Matching with Shape Contexts, 點選進入後可以看到簡單的介紹,

這個方法有點像是generalized houghline投票機制, 不過很可惜OpenCV目前還沒有納入

至於會不會就是上述提到的ShapeDistanceExtractor各位可以拭目以待(因為我也不知道)

後續: 寫完沒多久仔細看了一下線上手冊,

沒錯!!!ShapeDistanceExtractor就是實作以下這篇paper,感動~~~~

Implementation of the Shape Context descriptor and matching algorithm proposed by Belongie et al. in “Shape Matching and Object Recognition Using Shape Contexts” (PAMI 2002)

===================================================================================

來源資料:Matching with Shape Contexts

Background

The basic idea behind shape contexts is illustrated in the following diagrams. Given a set of points from an image (e.g. extracted from a set of detected edge elements), the shape context captures the relative distribution of points in the plane relative to each point on the shape. Specifically, we compute a histogram using log-polar coordinates as shown here:

Thus we get descriptors that are similar for homologous (corresponding) points and dissimilar for non-homologous points, as illustrated here, where the bin counts in the histogram are indicated by the grayshade (black=large, white=small):

We then use the shape contexts as attributes for a weighted bipartite matching problem.