Object Recognition Using Outline Images
M. Izumi, Kenji Kusakabe, Hiroshi Kawakami, K. Fukunaga · 2005
We propose a method of object recog- nition using the knowledge of the object world de- scribed in a form of a frame structure. The hier- archical frame structure consists of two levels, an outline image level and an aspect graph level. Us- ing this database, we can recognize input object as an object that has the largest degree of similar- ity between input edge image and outline images and/or aspect graphs in the database. Firstly, we calculate the degree of similarity between the in- put image and images in the database on the out- line level. As a result of this, we can narrow down the candidates and refine the extracted edge image to more precise edge line image on the aspect level. In this point, it is available to recognize objects on high recognition rates. In the case that the object world consists of five kinds of chairs composed by polygonal surfaces, we show some experimental re- sults.