The clustering of appearance of the object under the illumination condition for changing

摘要

[0060] removing gathering of the picture which does not have the label from midst of the collection of the object which under the image pickup conditions for differing is acquired, the element it disassembles that gathering mutually corresponds to the individual object in subclass, the clustering is necessary. The illumination condition for changing (100a) gathering of the picture of the 3D object which is acquired with under (101c) as for appearance based method in order the clustering to do, it is possible to be based on the concept of the illumination gimlet. Problem of the clustering although the convex versatile gimlet (301c) you find regarding the image space of high dimension is equal. Probability of the picture of the opposite which belongs to the versatile gimlet which in order to judge the gimlet structure which hides in the graphics data efficiently, it is possible, becomes the same cause to use the concept similarity of the gimlet, is measured. As for other algorithm, it is possible to be based on the similarity measurement which is made on the basis of the picture slope comparison which processes direct image slope with the size of picture slope and comparing direction.